├── README.md
├── chk.py
├── cnv_model.py
├── dist.py
├── ds_load.py
├── iam
    ├── iam.gin
    ├── iam.sh
    ├── iam_ln.gin
    ├── iam_par_gt.py
    ├── test.gc
    ├── test.ln
    ├── train.gc
    ├── train.ln
    ├── val.gc
    └── val.ln
├── ich17
    ├── ich.gin
    ├── ich.py
    ├── ich.sh
    ├── test
    └── train
├── o1.png
├── requirements.txt
├── test.py
├── train.py
└── utils.py


/README.md:
--------------------------------------------------------------------------------
  1 | # OrigamiNet
  2 | 
  3 | Public implementation of our CVPR 2020 paper:
  4 | 
  5 | "OrigamiNet: Weakly-Supervised, Segmentation-Free, One-Step, Full Page TextRecognition by learning to unfold"
  6 | 
  7 | <p align="center">
  8 | <img src="./o1.png">
  9 | </p>
 10 | 
 11 | ## Getting Started
 12 | 
 13 | OrigamiNet has been implemented and tested with Python 3.6 and PyTorch 1.3. All project configuration is handled using [Gin](https://github.com/google/gin-config).
 14 | 
 15 | First clone the repo:
 16 | ```
 17 | git clone https://github.com/IntuitionMachines/OrigamiNet.git
 18 | ```
 19 | Then install the dependencies with:
 20 | ```
 21 | pip install -r requirements.txt
 22 | ```
 23 | ## Replicating Experiments
 24 | 
 25 | ### IAM
 26 | 1. Register at the FKI's webpage [here](http://www.fki.inf.unibe.ch/DBs/iamDB/iLogin/index.php).
 27 | 
 28 | 2. After obtaining the username and password, we provide a script to download and setup the dataset, crop paragraph images and generate corresponding paragraph transcriptions by concatenating each line transcription. Run:
 29 | ```
 30 | bash iam/iam.sh $IAM_USER $IAM_PASS $IAM_DEST
 31 | ```
 32 | where `$IAM_USER` and `$IAM_PASS` are the username and password from FKI website, `IAM_DEST` is the destination folder where the dataset will be saved (the folder will be created by the script if it doesn't exist).
 33 | 
 34 | 3. Run the training script using provided configuration:
 35 | ```
 36 | python train.py --gin iam/iam.gin
 37 | ```
 38 | Note: if you want to use `horovod`, run as following:
 39 | ```
 40 | horovodrun -n $N_GPU -H localhost:$N_GPU python train.py --gin iam/iam.gin
 41 | ```
 42 | where `$N_GPU` is the number of gpus to be used (visible GPUs can be controlled by setting `CUDA_VISIBLE_DEVICES`)
 43 | 
 44 | ### ICDAR2017 HTR
 45 | 
 46 | 1. Download and set up the dataset using the provided script:
 47 | ```
 48 | bash ich17/ich.sh $ICH_DEST
 49 | ```
 50 | `ICH_DEST` is the destination folder where the dataset will be saved. The folder will be created by the script if it doesn't exist.
 51 | 
 52 | 2. Run the training script using provided configuration:
 53 | ```
 54 | python train.py --gin ich17/ich.gin
 55 | ```
 56 | 
 57 | ## Results
 58 | 
 59 | In the following table CER and nCER are respectively the micro and macro averaged [Character Error Rate](https://en.wikipedia.org/wiki/Levenshtein_distance). BLEU is the marco-averaged character-level [BLEU](https://en.wikipedia.org/wiki/BLEU) score.
 60 | 
 61 | ### Paper results
 62 | 
 63 | Dataset | wmul | Size    | CER (%)  | nCER (%) | BLEU
 64 | --------|------|---------|--------- | -------- |-----
 65 | IAM     | 1.5  | 750x750 |  4.7     | 4.84     | 91.15
 66 | ICDAR   | 1.8  |1400x1000|  6.80    | 5.87     | 92.67
 67 | 
 68 | 
 69 | ### Additional results
 70 | 
 71 | Dataset | wmul | Size    | CER (%)  | nCER (%) | BLEU
 72 | --------|------|---------|--------- | -------- |-----
 73 | IAM     | 1.0  | 750x750 |  4.85    | 4.95     | 90.87
 74 | IAM     | 2.0  | 750x750 |  4.41    | 4.54     | 91.25
 75 | IAM     | 3.0  | 750x750 |  4.29    | 4.41     | 91.84
 76 | IAM     | 4.0  | 750x750 |  4.07    | 4.18     | 92.21
 77 | ICDAR   | 2.4  |1400x1000|  6.01    | 5.30     | 93.64
 78 | 
 79 | These experiments were done with a `batch_size` of 8. We also obtained promising results with a `batch_size` of 4, as the proposed architecure does not utilize BatchNorm operations.
 80 | 
 81 | ### Synthetic hard-to-segment IAM variants
 82 | 
 83 | In the paper, two IAM variants with hard-to-segment text-lines were presented. These results can be replicated as follows:
 84 | 
 85 | #### Compact lines
 86 | 
 87 | 1. Make a copy of the `pargs` folder, which contains the extracted paragraph images:
 88 | ```
 89 | cp -r iam/pargs/ iam/pargsCL
 90 | ```
 91 | 2. To generate IAM with touching lines, use `image-magick` to resize images to half the height using seam carving.
 92 | 
 93 | The following line runs the conversion in parallel to speed up the process:
 94 | ```
 95 | find iam/pargsCL -iname "*.png" -type f -print0 | parallel --progress -0 -j +0 "mogrify -liquid-rescale 100x50%\! {}"
 96 | ```
 97 | 
 98 | #### Rotated and warped
 99 | 
100 | 1. Make a copy of the `pargs` folder, which contains the extracted paragraph images:
101 | ```
102 | cp -r iam/pargs/ iam/pargsPW
103 | ```
104 | 2. To generate IAM with a random projection and wavy text-lines:
105 | 
106 | ```
107 | find iam/pargsPW -iname "*.png" -type f -print0 | parallel --progress -0 -j +0 "python dist.py {}"
108 | ```
109 | ### Results
110 | 
111 | Dataset            | wmul | Size    | CER (%)  
112 | -------------------|------|---------|--------- 
113 | Compact lines      | 1.0  | 750x750 |  6.0    
114 | Rotated and warped | 1.0  | 750x750 |  5.6    
115 | 
116 | ## Single line results
117 | 
118 | To be as useful as possible, we show how to perform single-line recognition based on the code. This essentially resembles the [GTR model](https://arxiv.org/abs/1812.11894). Assuming lines from IAM and thier transcriptions are stored in `iam/lines/`, run as
119 | ```
120 | python train.py --gin iam/iam_ln.gin
121 | ```
122 | ### Results
123 | 
124 | Results on the IAM single-line test set
125 | 
126 | Dataset        | nlyrs | Size    | CER (%)  
127 | ---------------|-------|---------|--------- 
128 | IAM lines      | 12    | 32x600  |  5.26    
129 | IAM lines      | 18    | 32x600  |  4.84    
130 | IAM lines      | 24    | 32x600  |  4.76     
131 | 
132 | 
133 | 
134 | ## Gin Options
135 | 
136 | This is a brief list of the most important gin options. For full config files see `iam/iam.gin` or `ich17/ich.gin`
137 | - `dist`: The parallel traning method. We currently support three possible values: 
138 |   - `DP` uses DataParallel
139 |   - `DDP` uses DistributedDataParallel
140 |   - `HVD` uses horovod
141 | - `n_channels`: number of channels per image
142 | - `o_classes`: The size of the target vocabulary (i.e. number of symbols in the alphabet)
143 | - `GradCheck`: Whether or not to use gradient checkpointing
144 |   - `0` disabled
145 |   - `1` enabled, light version which offers good memory saving with small slowdown
146 |   - `2` enabled, higher memory saving, but noticeably slower than `1`
147 | - `get_images.max_h` and `get_images.max_w`: Target height and width for each image, images will be resized to this target dimentions while maintaining aspect ratio by padding.
148 | - `train.AMP`: Whether Automatic Mixed Precision (by [Nvidia apex](https://github.com/NVIDIA/apex)) is enabled
149 | - `train.WdB`: Whether [Wandb](https://github.com/wandb/client) logging is enabled
150 | - `train.train_data_list` and `train.test_data_list`: Path to file containing list of training or testing images
151 | - `train.train_data_path` and `train.test_data_path`: Path to folder containing the training or testing images
152 | - `train.train_batch_size` and `train.val_batch_size`: the batch size used during training and validation, the interpretation of this value vaires according to `dist` option
153 |   - `DP` the `train.batch_size` is the total batch size
154 |   - `DDP` or `HVD` then `train.batch_size` is the batch size per process (total batch size is `train.batch_size*#Processes`
155 | - `train.workers`: Number of worker for the PyTorch `DataLoader`
156 | - `train.continue_model`: Path to checkpoint to continue from
157 | - `train.num_iter`: Total number of training iterations
158 | - `train.valInterval`: Perform validation every how many batches
159 | - `OrigamiNet.nlyrs`: #layers in the GTR model
160 | - `OrigamiNet.reduceAxis`: Final axis of reduction
161 | - `OrigamiNet.wmul`: Channel multipler, numer of channels in each channel will be multiplied by this value
162 | - `OrigamiNet.lszs`: Number of channels for each layer, this is a dictionary of format `leyer_id:channels`, unspecified layers are assumed constant
163 | - `s1/Upsample.size`: Size of penultimate layer
164 | - `s1/Upsample.size`: Size of the last layer
165 | - `OrigamiNet.lreszs`: Resampling stages in the model
166 | 
167 | ## Acknowledgements
168 | 
169 | Some code is borrowed from the [deep-text-recognition-benchmark](https://github.com/clovaai/deep-text-recognition-benchmark),
170 | which is under the Apache 2.0 license.
171 | 
172 | Network architecture was visualized using [PlotNeuralNet](https://github.com/HarisIqbal88/PlotNeuralNet)
173 | 
174 | This work was sponsored by [Intuition Machines, Inc](https://www.imachines.com/).
175 | 
176 | ## Citation
177 | 
178 | ```bibtex
179 | @inproceedings{yousef2020origaminet,
180 |   title={OrigamiNet: Weakly-Supervised, Segmentation-Free, One-Step, Full Page TextRecognition by learning to unfold},
181 |   author={Yousef, Mohamed and Bishop, Tom E.},
182 |   booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
183 |   month = {June},
184 |   year = {2020}
185 | }
186 | ```
187 | 


--------------------------------------------------------------------------------
/chk.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | import warnings
 3 | 
 4 | from torch.utils.checkpoint import checkpoint
 5 | 
 6 | 
 7 | def checkpoint_sequential_step(functions, segments, *inputs, **kwargs):
 8 |     r"""A helper function for checkpointing sequential models.
 9 | 
10 |     Sequential models execute a list of modules/functions in order
11 |     (sequentially). Therefore, we can divide such a model in various segments
12 |     and checkpoint each segment. All segments except the last will run in
13 |     :func:`torch.no_grad` manner, i.e., not storing the intermediate
14 |     activations. The inputs of each checkpointed segment will be saved for
15 |     re-running the segment in the backward pass.
16 | 
17 |     See :func:`~torch.utils.checkpoint.checkpoint` on how checkpointing works.
18 | 
19 |     .. warning::
20 |         Checkpointing doesn't work with :func:`torch.autograd.grad`, but only
21 |         with :func:`torch.autograd.backward`.
22 | 
23 |     .. warning:
24 |         At least one of the inputs needs to have :code:`requires_grad=True` if
25 |         grads are needed for model inputs, otherwise the checkpointed part of the
26 |         model won't have gradients.
27 | 
28 |     Args:
29 |         functions: A :class:`torch.nn.Sequential` or the list of modules or
30 |             functions (comprising the model) to run sequentially.
31 |         segments: Number of chunks to create in the model
32 |         inputs: tuple of Tensors that are inputs to :attr:`functions`
33 |         preserve_rng_state(bool, optional, default=True):  Omit stashing and restoring
34 |             the RNG state during each checkpoint.
35 | 
36 |     Returns:
37 |         Output of running :attr:`functions` sequentially on :attr:`*inputs`
38 | 
39 |     Example:
40 |         >>> model = nn.Sequential(...)
41 |         >>> input_var = checkpoint_sequential(model, chunks, input_var)
42 |     """
43 |     # Hack to mix *args with **kwargs in a python 2.7-compliant way
44 |     preserve = kwargs.pop('preserve_rng_state', True)
45 |     if kwargs:
46 |         raise ValueError("Unexpected keyword arguments: " + ",".join(arg for arg in kwargs))
47 | 
48 |     # To accept variadic arguments is not consistent with nn.Sequential.
49 |     # This interface will be changed at PyTorch 1.3.
50 |     # See also: https://github.com/pytorch/pytorch/issues/19260
51 |     if not inputs:
52 |         warnings.warn('Giving no input to checkpoint_sequential has been deprecated, '
53 |                       'a TypeError will be raised after PyTorch 1.3',
54 |                       DeprecationWarning)
55 |     elif len(inputs) > 1:
56 |         warnings.warn('multiple inputs to checkpoint_sequential has been deprecated, '
57 |                       'a TypeError will be raised after PyTorch 1.3',
58 |                       DeprecationWarning)
59 | 
60 |     def run_function(start, end, functions):
61 |         def forward(*inputs):
62 |             for j in range(start, end + 1):
63 |                 if isinstance(inputs, tuple):
64 |                     inputs = functions[j](*inputs)
65 |                 else:
66 |                     inputs = functions[j](inputs)
67 |             return inputs
68 |         return forward
69 | 
70 |     if isinstance(functions, torch.nn.Sequential):
71 |         functions = list(functions.children())
72 | 
73 |     segment_size = segments
74 |     # the last chunk has to be non-volatile
75 |     end = -1
76 |     for start in range(0, len(functions)-segments, segments):
77 |         end = start + segment_size - 1
78 |         inputs = checkpoint(run_function(start, end, functions), *inputs,
79 |                             preserve_rng_state=preserve)
80 |         if not isinstance(inputs, tuple):
81 |             inputs = (inputs,)
82 | 
83 |     return checkpoint(run_function(end + 1, len(functions) - 1, functions), *inputs)
84 | 


--------------------------------------------------------------------------------
/cnv_model.py:
--------------------------------------------------------------------------------
  1 | import torch
  2 | import torch.nn as nn
  3 | import torch.nn.functional as F
  4 | from chk import checkpoint_sequential_step, checkpoint
  5 | 
  6 | import math
  7 | import numpy as np
  8 | from torchvision.utils import save_image
  9 | 
 10 | import gin
 11 | 
 12 | def ginM(n): return gin.query_parameter(f'%{n}')
 13 | gin.external_configurable(nn.MaxPool2d, module='nn')
 14 | gin.external_configurable(nn.Upsample,  module='nn')
 15 | 
 16 | 
 17 | class LN(nn.Module):
 18 |     def forward(self, x):
 19 |         return F.layer_norm(x, x.size()[1:], weight=None, bias=None, eps=1e-05)
 20 | 
 21 | @gin.configurable
 22 | class PadPool(nn.Module):
 23 |     def forward(self, x):
 24 |         x = F.pad(x, [0, 0, 0, 1]) 
 25 |         x = F.max_pool2d(x,(2, 2), stride=(1, 2))
 26 |         return x
 27 | 
 28 | def pCnv(inp,out,groups=1):
 29 |   return nn.Sequential(
 30 |       nn.Conv2d(inp,out,1,bias=False,groups=groups),
 31 |       nn.InstanceNorm2d(out,affine=True)
 32 |   )
 33 |   
 34 | #regarding same padding in PT https://github.com/pytorch/pytorch/issues/3867
 35 | def dsCnv(inp,k):
 36 |   return nn.Sequential(
 37 |       nn.Conv2d(inp,inp,k,groups=inp,bias=False,padding=(k - 1) // 2),
 38 |       nn.InstanceNorm2d(inp,affine=True)
 39 |   )
 40 | 
 41 | ngates = 2
 42 | 
 43 | class Gate(nn.Module):
 44 |     def __init__(self,ifsz):
 45 |         super().__init__()
 46 |         self.ln = LN()
 47 | 
 48 |     def forward(self, x):
 49 |         t0,t1 = torch.chunk(x, ngates, dim=1)
 50 |         t0 = torch.tanh_(t0)
 51 |         t1.sub_(2)
 52 |         t1 = torch.sigmoid_(t1)
 53 | 
 54 |         return t1*t0
 55 | 
 56 | def customGC(module):
 57 |     def custom_forward(*inputs):
 58 |         inputs = module(inputs[0])
 59 |         return inputs
 60 |     return custom_forward
 61 | 
 62 | @gin.configurable
 63 | class GateBlock(nn.Module):
 64 |     def __init__(self, ifsz, ofsz, gt = True, ksz = 3, GradCheck=gin.REQUIRED):
 65 |         super().__init__()
 66 | 
 67 |         cfsz   = int( math.floor(ifsz/2) )
 68 |         ifsz2  = ifsz + ifsz%2
 69 | 
 70 |         self.sq = nn.Sequential(
 71 |           pCnv(ifsz, cfsz),
 72 |           dsCnv(cfsz,ksz),
 73 |           nn.ELU(),
 74 |           ###########
 75 |           pCnv(cfsz, cfsz*ngates),
 76 |           dsCnv(cfsz*ngates,ksz),
 77 |           Gate(cfsz),
 78 |           ###########
 79 |           pCnv(cfsz, ifsz),
 80 |           dsCnv(ifsz,ksz),
 81 |           nn.ELU(),
 82 |         )
 83 | 
 84 |         self.gt = gt
 85 |         self.gc = GradCheck
 86 |     
 87 | 
 88 |     def forward(self, x):
 89 |         if self.gc >= 1:
 90 |           y = checkpoint(customGC(self.sq), x)
 91 |         else:
 92 |           y = self.sq(x)
 93 | 
 94 |         out = x + y
 95 |         return out
 96 | 
 97 | @gin.configurable
 98 | class InitBlock(nn.Module):
 99 |     def __init__(self, fup, n_channels):
100 |         super().__init__()
101 | 
102 |         self.n1 = LN()
103 |         self.Initsq = nn.Sequential(
104 |           pCnv(n_channels, fup),
105 |           nn.Softmax(dim=1),
106 |           dsCnv(fup,11),
107 |           LN()
108 |         )
109 | 
110 |     def forward(self, x):
111 |         x  = self.n1(x)
112 |         xt = x
113 |         x  = self.Initsq(x)
114 |         x  = torch.cat([x,xt],1)
115 |         return x
116 | 
117 | @gin.configurable
118 | class OrigamiNet(nn.Module):
119 |     def __init__(self, n_channels, o_classes, wmul, lreszs, lszs, nlyrs, fup, GradCheck, reduceAxis=3):
120 |         super().__init__()
121 | 
122 |         self.lreszs = lreszs
123 |         self.Initsq = InitBlock(fup)
124 |         
125 |         layers = []
126 |         isz = fup + n_channels
127 |         osz = isz
128 |         for i in range(nlyrs):
129 |             osz = int( math.floor(lszs[i] * wmul) ) if i in lszs else isz    
130 |             layers.append( GateBlock(isz, osz, True, 3) )
131 | 
132 |             if isz != osz:
133 |               layers.append( pCnv(isz, osz) )
134 |               layers.append( nn.ELU() )
135 |             isz = osz
136 | 
137 |             if i in lreszs:
138 |               layers.append( lreszs[i] )
139 |  
140 |         layers.append( LN() )
141 |         self.Gatesq = nn.Sequential(*layers)
142 |         
143 |         self.Finsq = nn.Sequential(
144 |           pCnv(osz, o_classes),
145 |           nn.ELU(),
146 |         )
147 | 
148 |         self.n1 = LN()
149 |         self.it=0
150 |         self.gc = GradCheck
151 |         self.reduceAxis = reduceAxis
152 | 
153 |     def forward(self, x, t=[]):
154 |         x = self.Initsq(x)
155 | 
156 |         if self.gc >=2:
157 |           x = checkpoint_sequential_step(self.Gatesq,4,x)  #slower, more memory save
158 |           # x = checkpoint_sequential_step(self.Gatesq,8,x)  #faster, less memory save
159 |         else:
160 |           x = self.Gatesq(x)
161 | 
162 |         x = self.Finsq(x)
163 |  
164 |         x = torch.mean(x, self.reduceAxis, keepdim=False)
165 |         x = self.n1(x)
166 |         x = x.permute(0,2,1)
167 | 
168 |         return x


--------------------------------------------------------------------------------
/dist.py:
--------------------------------------------------------------------------------
  1 | import numpy as np
  2 | import torch
  3 | from scipy import ndimage
  4 | import scipy.misc
  5 | import skimage
  6 | from skimage import exposure
  7 | import matplotlib.pyplot as plt
  8 | import os,sys
  9 | from collections import namedtuple
 10 | from random import shuffle
 11 | import itertools
 12 | from sklearn.externals import joblib
 13 | from skimage import transform as stf
 14 | from PIL import Image
 15 | import unicodedata
 16 | import copy
 17 | from torch.utils.data import Dataset
 18 | import imageio
 19 | from math import floor, ceil
 20 | import random
 21 | 
 22 | try:
 23 |    import cPickle as pickle
 24 | except:
 25 |    import pickle
 26 | 
 27 | # Based on Elastic distortions in
 28 | # https://github.com/mdbloice/Augmentor/blob/master/Augmentor/Operations.py
 29 | class Distort3():
 30 | 
 31 |     def __init__(self, probability, grid_width, grid_height, magnitudeX, magnitudeY, Isize, min_h_sep, min_v_sep):
 32 | 
 33 |         self.probability = probability
 34 |         self.grid_width = grid_width
 35 |         self.grid_height = grid_height
 36 |         self.xmagnitude = abs(magnitudeX)
 37 |         self.ymagnitude = abs(magnitudeY)
 38 |         self.randomise_magnitude = True
 39 | 
 40 |         w, h = Isize
 41 | 
 42 |         horizontal_tiles = self.grid_width
 43 |         vertical_tiles = self.grid_height
 44 | 
 45 |         width_of_square = int(floor(w / float(horizontal_tiles)))
 46 |         height_of_square = int(floor(h / float(vertical_tiles)))
 47 | 
 48 |         width_of_last_square = w - (width_of_square * (horizontal_tiles - 1))
 49 |         height_of_last_square = h - (height_of_square * (vertical_tiles - 1))
 50 | 
 51 |         dimensions = []
 52 |         shift = [[(0,0) for x in range(horizontal_tiles)] for y in range(vertical_tiles)]
 53 | 
 54 |         for vertical_tile in range(vertical_tiles):
 55 |             for horizontal_tile in range(horizontal_tiles):
 56 |                 if vertical_tile == (vertical_tiles - 1) and horizontal_tile == (horizontal_tiles - 1):
 57 |                     dimensions.append([horizontal_tile * width_of_square,
 58 |                                        vertical_tile * height_of_square,
 59 |                                        width_of_last_square + (horizontal_tile * width_of_square),
 60 |                                        height_of_last_square + (height_of_square * vertical_tile)])
 61 |                 elif vertical_tile == (vertical_tiles - 1):
 62 |                     dimensions.append([horizontal_tile * width_of_square,
 63 |                                        vertical_tile * height_of_square,
 64 |                                        width_of_square + (horizontal_tile * width_of_square),
 65 |                                        height_of_last_square + (height_of_square * vertical_tile)])
 66 |                 elif horizontal_tile == (horizontal_tiles - 1):
 67 |                     dimensions.append([horizontal_tile * width_of_square,
 68 |                                        vertical_tile * height_of_square,
 69 |                                        width_of_last_square + (horizontal_tile * width_of_square),
 70 |                                        height_of_square + (height_of_square * vertical_tile)])
 71 |                 else:
 72 |                     dimensions.append([horizontal_tile * width_of_square,
 73 |                                        vertical_tile * height_of_square,
 74 |                                        width_of_square + (horizontal_tile * width_of_square),
 75 |                                        height_of_square + (height_of_square * vertical_tile)])
 76 | 
 77 |                 sm_h = min( self.xmagnitude, width_of_square  - (min_h_sep+shift[vertical_tile][horizontal_tile-1][0])  ) if horizontal_tile>0 else self.xmagnitude
 78 |                 sm_v = min( self.ymagnitude, height_of_square - (min_v_sep+shift[vertical_tile-1][horizontal_tile][1])  ) if vertical_tile>0   else self.ymagnitude
 79 | 
 80 |                 dx = random.randint(-sm_h, self.xmagnitude)
 81 |                 dy = random.randint(-sm_v, self.ymagnitude)
 82 |                 shift[vertical_tile][horizontal_tile] = (dx,dy)
 83 | 
 84 | 
 85 |         shift = list(itertools.chain.from_iterable(shift))
 86 | 
 87 |         
 88 |         last_column = []
 89 |         for i in range(vertical_tiles):
 90 |             last_column.append((horizontal_tiles - 1) + horizontal_tiles * i)
 91 | 
 92 |         last_row = range((horizontal_tiles * vertical_tiles) - horizontal_tiles, horizontal_tiles * vertical_tiles)
 93 | 
 94 |         polygons = []
 95 |         for x1, y1, x2, y2 in dimensions:
 96 |             polygons.append([x1, y1, x1, y2, x2, y2, x2, y1])
 97 | 
 98 |         polygon_indices = []
 99 |         for i in range((vertical_tiles * horizontal_tiles) - 1):
100 |             if i not in last_row and i not in last_column:
101 |                 polygon_indices.append([i, i + 1, i + horizontal_tiles, i + 1 + horizontal_tiles])
102 | 
103 |         for id,(a, b, c, d) in enumerate(polygon_indices):
104 |         
105 |             dx = shift[id][0]
106 |             dy = shift[id][1]
107 | 
108 |             x1, y1, x2, y2, x3, y3, x4, y4 = polygons[a]
109 |             polygons[a] = [x1, y1,
110 |                            x2, y2,
111 |                            x3 + dx, y3 + dy,
112 |                            x4, y4]
113 | 
114 |             x1, y1, x2, y2, x3, y3, x4, y4 = polygons[b]
115 |             polygons[b] = [x1, y1,
116 |                            x2 + dx, y2 + dy,
117 |                            x3, y3,
118 |                            x4, y4]
119 | 
120 |             x1, y1, x2, y2, x3, y3, x4, y4 = polygons[c]
121 |             polygons[c] = [x1, y1,
122 |                            x2, y2,
123 |                            x3, y3,
124 |                            x4 + dx, y4 + dy]
125 | 
126 |             x1, y1, x2, y2, x3, y3, x4, y4 = polygons[d]
127 |             polygons[d] = [x1 + dx, y1 + dy,
128 |                            x2, y2,
129 |                            x3, y3,
130 |                            x4, y4]
131 | 
132 |         generated_mesh = []
133 |         for i in range(len(dimensions)):
134 |             generated_mesh.append([dimensions[i], polygons[i]])
135 | 
136 |         self.generated_mesh = generated_mesh
137 | 
138 |     def perform_operation(self, image):
139 |         return image.transform(image.size, Image.MESH, self.generated_mesh, resample=Image.BICUBIC)
140 | 
141 | 
142 | def aug_ED2(imgs, w, h, n_ch, tst=False):
143 | 
144 |     d = Distort3(1.0, 10, 10, 0, 25, [w, h], 1, 1)
145 | 
146 |     for i in range(len(imgs)):        
147 |         res = d.perform_operation(Image.fromarray(np.squeeze((imgs[i] * 255).astype(np.uint8))))
148 |         imgs[i] = np.reshape(res, [h, w, n_ch])
149 | 
150 |     return np.squeeze(imgs)
151 | 
152 | 
153 | def RndTform(img,val=125 * 1.5):
154 | 
155 |     Ih,Iw = img[0].shape[:2]
156 | 
157 |     sgn = torch.randint(0,2,(1,)).item() * 2 - 1
158 | 
159 |     if sgn>0:
160 |         dw = val
161 |         dh = 0
162 |     else:
163 |         dw = 0
164 |         dh = val
165 | 
166 |     def rd(d): return torch.empty(1).uniform_(-d,d).item()
167 |     def fd(d): return torch.empty(1).uniform_(-dw,d).item()
168 | 
169 |     tl_top = rd(dh)  # Top left corner, top margin
170 |     tl_left = fd(dw)  # Top left corner, left margin
171 |     bl_bottom = rd(dh)  # Bottom left corner, bottom margin
172 |     bl_left = fd(dw)  # Bottom left corner, left margin
173 |     tr_top = rd(dh)  # Top right corner, top margin
174 |     tr_right = fd( min(Iw * 3/4 - tl_left,dw) )  # Top right corner, right margin
175 |     br_bottom = rd(dh)  # Bottom right corner, bottom margin
176 |     br_right = fd( min(Iw * 3/4 - bl_left,dw) )  # Bottom right corner, right margin
177 |     
178 | 
179 |     tform = stf.ProjectiveTransform()
180 | 
181 |     tform.estimate(np.array((
182 |         (tl_left, tl_top),
183 |         (bl_left, Ih - bl_bottom),
184 |         (Iw - br_right, Ih - br_bottom),
185 |         (Iw - tr_right, tr_top)
186 |     )), np.array((
187 |         [0, 0 ],
188 |         [0, Ih - 1 ],
189 |         [Iw-1, Ih-1 ],
190 |         [Iw-1, 0]
191 |     )))
192 | 
193 |     corners = np.array([
194 |         [0, 0 ],
195 |         [0, Ih - 1 ],
196 |         [Iw-1, Ih-1 ],
197 |         [Iw-1, 0]
198 |     ])
199 | 
200 |     corners = tform.inverse(corners)
201 | 
202 |     minc = corners[:, 0].min()
203 |     minr = corners[:, 1].min()
204 |     maxc = corners[:, 0].max()
205 |     maxr = corners[:, 1].max()
206 |     out_rows = maxr - minr + 1
207 |     out_cols = maxc - minc + 1
208 |     output_shape = np.around((out_rows, out_cols))
209 |     
210 |     translation = (minc, minr)
211 |     tform4 = stf.SimilarityTransform(translation=translation)
212 |     tform = tform4 + tform
213 |     tform.params /= tform.params[2, 2]
214 |     
215 |     ret = []
216 |     for i in range(len(img)):
217 |         img2 = stf.warp(img[i], tform, output_shape=output_shape, cval=1.0)
218 |         img2 = stf.resize(img2, (Ih,Iw), preserve_range=True).astype(np.float32)
219 |         ret.append(img2)
220 | 
221 |     return ret
222 | 
223 | def npThum(img, max_w, max_h):
224 |     x, y = np.shape(img)[:2]
225 | 
226 |     y = min(int( y * max_h / x ),max_w)
227 |     x = max_h
228 | 
229 |     img = np.array(Image.fromarray(img).resize((y,x)))
230 |     return img
231 | 
232 | image_data = np.array(Image.open(sys.argv[1]))
233 | 
234 | image_data = npThum(image_data, 750, 750)
235 | image_data = skimage.img_as_float32(image_data)
236 | if image_data.ndim < 3:
237 |     image_data = np.expand_dims(image_data, axis=-1)
238 | 
239 | images = image_data[None,...]
240 | 
241 | images = np.array([RndTform([image], val=140)[0] for image in images])
242 | sh = images.shape
243 | images = np.array([aug_ED2(image[None,...],sh[2], sh[1], sh[3], tst=False) for image in images])
244 | imageio.imwrite(sys.argv[1],images[0].astype(np.uint8))
245 | 


--------------------------------------------------------------------------------
/ds_load.py:
--------------------------------------------------------------------------------
  1 | import numpy as np
  2 | import torch
  3 | from scipy import ndimage
  4 | import scipy.misc
  5 | import skimage
  6 | import os,sys
  7 | import itertools
  8 | from skimage import transform as stf
  9 | from PIL import Image
 10 | from torch.utils.data import Dataset
 11 | import imageio
 12 | from math import floor, ceil
 13 | import pickle
 14 | import gin
 15 | 
 16 | def RndTform(img,val):
 17 |     Ih,Iw = img[0].shape[:2]
 18 |     
 19 |     sgn = torch.randint(0,2,(1,)).item() * 2 - 1
 20 | 
 21 |     if sgn>0:
 22 |         dw = val
 23 |         dh = 0
 24 |     else:
 25 |         dw = 0
 26 |         dh = val
 27 | 
 28 |     def rd(d): return torch.empty(1).uniform_(-d,d).item()
 29 |     def fd(d): return torch.empty(1).uniform_(-dw,d).item()
 30 | 
 31 |     # generate a random projective transform
 32 |     # adapted from https://navoshta.com/traffic-signs-classification/
 33 |     tl_top = rd(dh)
 34 |     tl_left = fd(dw)
 35 |     bl_bottom = rd(dh)
 36 |     bl_left = fd(dw)
 37 |     tr_top = rd(dh)
 38 |     tr_right = fd( min(Iw * 3/4 - tl_left,dw) )
 39 |     br_bottom = rd(dh)
 40 |     br_right = fd( min(Iw * 3/4 - bl_left,dw) )
 41 | 
 42 |     tform = stf.ProjectiveTransform()
 43 |     tform.estimate(np.array((
 44 |         (tl_left, tl_top),
 45 |         (bl_left, Ih - bl_bottom),
 46 |         (Iw - br_right, Ih - br_bottom),
 47 |         (Iw - tr_right, tr_top)
 48 |     )), np.array((
 49 |         [0, 0 ],
 50 |         [0, Ih - 1 ],
 51 |         [Iw-1, Ih-1 ],
 52 |         [Iw-1, 0]
 53 |     )))
 54 | 
 55 |     # determine shape of output image, to preserve size
 56 |     # trick take from the implementation of skimage.transform.rotate
 57 |     corners = np.array([
 58 |         [0, 0 ],
 59 |         [0, Ih - 1 ],
 60 |         [Iw-1, Ih-1 ],
 61 |         [Iw-1, 0]
 62 |     ])
 63 | 
 64 |     corners = tform.inverse(corners)
 65 |     minc = corners[:, 0].min()
 66 |     minr = corners[:, 1].min()
 67 |     maxc = corners[:, 0].max()
 68 |     maxr = corners[:, 1].max()
 69 |     out_rows = maxr - minr + 1
 70 |     out_cols = maxc - minc + 1
 71 |     output_shape = np.around((out_rows, out_cols))
 72 | 
 73 |     # fit output image in new shape
 74 |     translation = (minc, minr)
 75 |     tform4 = stf.SimilarityTransform(translation=translation)
 76 |     tform = tform4 + tform
 77 |     # normalize
 78 |     tform.params /= tform.params[2, 2]
 79 |     
 80 | 
 81 |     ret = []
 82 |     for i in range(len(img)):
 83 |         img2 = stf.warp(img[i], tform, output_shape=output_shape, cval=1.0)
 84 |         img2 = stf.resize(img2, (Ih,Iw), preserve_range=True).astype(np.float32)
 85 |         ret.append(img2)
 86 | 
 87 | 
 88 |     return ret
 89 | 
 90 | @gin.configurable
 91 | def SameTrCollate(batch, prjAug, prjVal):
 92 |     images, labels = zip(*batch)
 93 | 
 94 |     images = [image.transpose((1,2,0)) for image in images]
 95 |     
 96 |     if prjAug:
 97 |         images = [RndTform([image], val=prjVal)[0] for image in images] #different transform to each batch
 98 |         # images = RndTform(images, val=prjVal) #apply same transform to all images in a batch
 99 |    
100 |     image_tensors = [torch.from_numpy(np.array(image, copy=False)) for image in images]
101 |     image_tensors = torch.cat([t.unsqueeze(0) for t in image_tensors], 0)
102 |     image_tensors = image_tensors.permute(0,3,1,2)
103 | 
104 |     return image_tensors, labels
105 | 
106 | 
107 | class myLoadDS(Dataset):
108 |     def __init__(self, flist, dpath, ralph=None, fmin=True, mln=None):
109 |         self.fns = get_files(flist, dpath)
110 |         self.tlbls = get_labels(self.fns)
111 |         
112 |         if ralph == None:
113 |             alph  = get_alphabet(self.tlbls)
114 |             self.ralph = dict (zip(alph.values(),alph.keys()))
115 |             self.alph = alph
116 |         else:
117 |             self.ralph = ralph
118 |         
119 |         if mln != None:
120 |             filt = [len(x) <= mln if fmin else len(x) >= mln for x in self.tlbls]
121 |             self.tlbls = np.asarray(self.tlbls)[filt].tolist()
122 |             self.fns   = np.asarray(self.fns  )[filt].tolist()
123 |     
124 |     def __len__(self):
125 |         return len(self.fns)
126 | 
127 |     def __getitem__(self, index):
128 |         timgs = get_images(self.fns[index])
129 |         timgs = timgs.transpose((2,0,1))
130 | 
131 |         return ( timgs , self.tlbls[index] )
132 | 
133 | def get_files(nfile, dpath):
134 |     fnames = open(nfile, 'r').readlines()
135 |     fnames = [ dpath + x.strip() for x in fnames ]
136 |     return fnames
137 | 
138 | def npThum(img, max_w, max_h):
139 |     x, y = np.shape(img)[:2]
140 | 
141 |     y = min(int( y * max_h / x ),max_w)
142 |     x = max_h
143 | 
144 |     img = np.array(Image.fromarray(img).resize((y,x)))
145 |     return img
146 | 
147 | @gin.configurable
148 | def get_images(fname, max_w, max_h, nch):
149 | 
150 |     try:
151 | 
152 |         image_data = np.array(Image.open(fname))
153 |         image_data = npThum(image_data, max_w, max_h)
154 |         image_data = skimage.img_as_float32(image_data)
155 | 
156 |         h, w = np.shape(image_data)[:2]
157 |         if image_data.ndim < 3:
158 |             image_data = np.expand_dims(image_data, axis=-1)
159 |         
160 |         if nch==3 and image_data.shape[2]!=3:
161 |             image_data = np.tile(image_data,3)
162 | 
163 |         image_data = np.pad(image_data,((0,0),(0,max_w-np.shape(image_data)[1]),(0,0)), mode='constant', constant_values=(1.0))
164 | 
165 |     except IOError as e:
166 |         print('Could not read:', fname, ':', e)
167 |     
168 |     return image_data
169 | 
170 | def get_labels(fnames):
171 | 
172 |     labels = []
173 |     for id,image_file in enumerate(fnames):
174 |         fn  = os.path.splitext(image_file)[0] + '.txt'
175 |         lbl = open(fn, 'r').read()
176 |         lbl = ' '.join(lbl.split()) #remove linebreaks if present
177 |         
178 |         labels.append(lbl)
179 | 
180 |     return labels
181 | 
182 | def get_alphabet(labels):
183 |     
184 |     coll = ''.join(labels)     
185 |     unq  = sorted(list(set(coll)))
186 |     unq  = [''.join(i) for i in itertools.product(unq, repeat = 1)]
187 |     alph = dict( zip( unq,range(len(unq)) ) )
188 | 
189 |     return alph


--------------------------------------------------------------------------------
/iam/iam.gin:
--------------------------------------------------------------------------------
 1 | port = 12345
 2 | manualSeed = 1111
 3 | dist = 'DDP'
 4 | 
 5 | n_channels = 1
 6 | o_classes = 80
 7 | 
 8 | GradCheck = 0
 9 | 
10 | # Parameters for get_images (ds_load.py):
11 | # ==============================================================================
12 | get_images.max_w = 750
13 | get_images.max_h = 750
14 | get_images.nch = %n_channels
15 | 
16 | # Parameters for SameTrCollate (ds_load.py):
17 | # ==============================================================================
18 | SameTrCollate.prjAug = True
19 | SameTrCollate.prjVal = 125
20 | 
21 | # Parameters for train (train.py):
22 | # ==============================================================================
23 | train.experiment_name = 'iam_gin_test_'
24 | 
25 | train.AMP = False
26 | train.WdB = False
27 | train.wdbprj = 'prj_Name'
28 | train.train_data_list = 'iam/train.gc'
29 | train.train_data_path = '/mnt/iam/pargs/'
30 | train.test_data_list  = 'iam/val.gc'
31 | train.test_data_path  = '/mnt/iam/pargs/'
32 | train.train_batch_size = 1
33 | train.val_batch_size   = 1
34 | train.workers = 4
35 | train.lr = 0.01
36 | train.continue_model = ''
37 | train.valInterval = 500
38 | train.num_iter = 0
39 | 
40 | 
41 | # Parameters for OrigamiNet (cnv_model.py):
42 | # ==============================================================================
43 | OrigamiNet.n_channels = %n_channels
44 | OrigamiNet.o_classes  = %o_classes
45 | OrigamiNet.fup = 33
46 | OrigamiNet.GradCheck = %GradCheck
47 | 
48 | OrigamiNet.nlyrs = 12
49 | OrigamiNet.wmul = 1.0
50 | 
51 | OrigamiNet.lszs = {0:  128,
52 |             2:  256,
53 |             4:  512,
54 |             11: 256  }
55 | 
56 | ##resize structure##
57 | MaxPool2d.kernel_size = (2,2)
58 | 
59 | Upsample.mode = 'bilinear'
60 | Upsample.align_corners = True
61 | s1/Upsample.size = (450,15)
62 | s2/Upsample.size = (1100,8)
63 | 
64 | OrigamiNet.lreszs = {
65 |                   0: @MaxPool2d(),
66 |                   2: @MaxPool2d(),
67 |                   4: @MaxPool2d(),
68 | 
69 |                   6: @PadPool(),
70 |                   8: @PadPool(),
71 |                   
72 |                   10: @s1/Upsample(),
73 |                   11: @s2/Upsample()
74 |             }
75 | 
76 | # Parameters for InitBlock (cnv_model.py):
77 | # ==============================================================================
78 | InitBlock.n_channels = %n_channels
79 | 
80 | # Parameters for GateBlock (cnv_model.py):
81 | # ==============================================================================
82 | GateBlock.GradCheck = %GradCheck


--------------------------------------------------------------------------------
/iam/iam.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | 
 3 | dst=$3
 4 | 
 5 | wget --user=$1 --password=$2 http://www.fki.inf.unibe.ch/DBs/iamDB/data/{forms/{formsA-D,formsE-H,formsI-Z}.tgz,xml/xml.tgz}
 6 | 
 7 | mkdir -p $dst/{forms,xml,pargs}
 8 | 
 9 | cat {formsA-D,formsE-H,formsI-Z}.tgz | tar -zxvf - -i -C $dst/forms
10 | tar zxvf xml.tgz -C $dst/xml
11 | 
12 | python `dirname "$0"`/iam_par_gt.py $dst


--------------------------------------------------------------------------------
/iam/iam_ln.gin:
--------------------------------------------------------------------------------
 1 | port = 12346
 2 | manualSeed = 1111
 3 | dist = 'DDP'
 4 | 
 5 | n_channels = 1
 6 | o_classes = 80
 7 | 
 8 | GradCheck = 0
 9 | 
10 | # Parameters for get_images (ds_load.py):
11 | # ==============================================================================
12 | get_images.max_w = 600
13 | get_images.max_h = 32
14 | get_images.nch = %n_channels
15 | 
16 | # Parameters for SameTrCollate (ds_load.py):
17 | # ==============================================================================
18 | SameTrCollate.prjAug = True
19 | SameTrCollate.prjVal = 16
20 | 
21 | # Parameters for train (train.py):
22 | # ==============================================================================
23 | train.experiment_name = 'iam_gin_lines_test_'
24 | 
25 | train.AMP = False
26 | train.WdB = False
27 | train.wdbprj = 'prj_Name'
28 | train.train_data_list = 'iam/train.ln'
29 | train.train_data_path = '/mnt/iam/lines/'
30 | train.test_data_list  = 'iam/val.ln'
31 | train.test_data_path  = '/mnt/iam/lines/'
32 | train.train_batch_size = 8
33 | train.val_batch_size   = 8
34 | train.workers = 4
35 | train.lr = 0.02
36 | train.continue_model = ''
37 | train.valInterval = 1000
38 | train.num_iter = 0
39 | 
40 | 
41 | # Parameters for OrigamiNet (cnv_model.py):
42 | # ==============================================================================
43 | OrigamiNet.n_channels = %n_channels
44 | OrigamiNet.o_classes  = %o_classes
45 | OrigamiNet.fup = 33
46 | OrigamiNet.GradCheck = %GradCheck
47 | OrigamiNet.reduceAxis = 2
48 | 
49 | OrigamiNet.nlyrs = 12
50 | OrigamiNet.wmul = 2.0
51 | 
52 | OrigamiNet.lszs = {0:  128,
53 |             2:  256,
54 |             4:  512,
55 |             6:  1024  
56 |             }
57 | 
58 | ##resize structure##
59 | MaxPool2d.kernel_size = (2,2)
60 | 
61 | OrigamiNet.lreszs = {
62 |                   0: @MaxPool2d(),
63 |                   2: @MaxPool2d(),
64 |             }
65 | 
66 | # Parameters for InitBlock (cnv_model.py):
67 | # ==============================================================================
68 | InitBlock.n_channels = %n_channels
69 | 
70 | # Parameters for GateBlock (cnv_model.py):
71 | # ==============================================================================
72 | GateBlock.GradCheck = %GradCheck


--------------------------------------------------------------------------------
/iam/iam_par_gt.py:
--------------------------------------------------------------------------------
 1 | import xml.etree.ElementTree as ET
 2 | from xml.sax.saxutils import unescape
 3 | import glob
 4 | import sys,os
 5 | import imageio
 6 | import os.path as osp
 7 | 
 8 | # base  = '/home/asd/'
 9 | base  = sys.argv[1]
10 | 
11 | xml   = osp.join(base, 'xml/')
12 | fpath = osp.join(base, 'forms/')
13 | gpath = osp.join(base, 'pargs/')
14 | 
15 | forms = glob.glob(xml+'*.xml')
16 | c=0
17 | for i in range(len(forms)):
18 |     print(forms[i])
19 |     rt = ET.parse(forms[i]).getroot()
20 |     xmin, ymin = [sys.maxsize] * 2
21 |     xmax, ymax = [0] * 2
22 |     ftxt = ''
23 |     for line in rt.findall('.//line'):
24 |         txt,id = line.attrib['text'],line.attrib['id']
25 |         txt = unescape(txt, {"&apos;": "'", "&quot;": '"'})
26 |         ftxt = ftxt + txt + '\n'
27 | 
28 |         for cmp in line.findall('.//cmp'):
29 |             dm = list(map(int,[ cmp.attrib['x'],cmp.attrib['y'],cmp.attrib['width'],cmp.attrib['height'] ]))
30 |             xmin = min(xmin,dm[0])
31 |             ymin = min(ymin,dm[1])
32 |             xmax = max(xmax,dm[0]+dm[2])
33 |             ymax = max(ymax,dm[1]+dm[3])
34 | 
35 |     ftxt = ftxt[:-1]
36 |     marg = 5
37 |     xmin -= marg
38 |     ymin -= marg
39 |     xmax += marg
40 |     ymax += marg
41 | 
42 |     frmid = os.path.splitext( os.path.basename(forms[i]) )[0]
43 |     frmfl = fpath + frmid + '.png'
44 |     frm = imageio.imread(frmfl)
45 |     imageio.imsave(gpath+frmid+'.png',frm[ymin:ymax,xmin:xmax])
46 |     open(gpath + frmid + '.txt','w').write(ftxt)
47 | 


--------------------------------------------------------------------------------
/iam/test.gc:
--------------------------------------------------------------------------------
  1 | c04-110.png
  2 | c04-116.png
  3 | c04-134.png
  4 | c04-139.png
  5 | c04-144.png
  6 | c04-150.png
  7 | c04-165.png
  8 | c04-170.png
  9 | c06-011.png
 10 | d01-016.png
 11 | d01-019.png
 12 | d01-049.png
 13 | d01-052.png
 14 | d01-080.png
 15 | d01-085.png
 16 | d01-098.png
 17 | d01-104.png
 18 | d01-118.png
 19 | d01-123.png
 20 | d03-117.png
 21 | d04-012.png
 22 | d04-016.png
 23 | d04-021.png
 24 | d04-028.png
 25 | d04-032.png
 26 | d04-037.png
 27 | d04-047.png
 28 | d04-050.png
 29 | d04-053.png
 30 | d04-058.png
 31 | d04-062.png
 32 | d04-066.png
 33 | d04-086.png
 34 | d04-089.png
 35 | d04-096.png
 36 | d04-101.png
 37 | d04-111.png
 38 | d04-125.png
 39 | d04-131.png
 40 | d05-008.png
 41 | d05-013.png
 42 | d05-021.png
 43 | d05-025.png
 44 | d05-030.png
 45 | d05-040.png
 46 | d06-003.png
 47 | d06-008.png
 48 | d06-011.png
 49 | d06-020.png
 50 | d06-025.png
 51 | d06-027.png
 52 | d06-037.png
 53 | d06-041.png
 54 | d06-046.png
 55 | d06-056.png
 56 | d06-060.png
 57 | d06-063.png
 58 | d06-067.png
 59 | d06-072.png
 60 | d06-076.png
 61 | d06-086.png
 62 | d06-096.png
 63 | d06-100.png
 64 | d06-104.png
 65 | d06-107.png
 66 | d06-111.png
 67 | d06-113.png
 68 | d07-082.png
 69 | d07-085.png
 70 | d07-089.png
 71 | d07-093.png
 72 | d07-096.png
 73 | d07-100.png
 74 | d07-102.png
 75 | e01-055.png
 76 | e06-000.png
 77 | e06-003.png
 78 | e06-010.png
 79 | e06-015.png
 80 | e06-021.png
 81 | e06-026.png
 82 | e06-030.png
 83 | e06-033.png
 84 | e06-046.png
 85 | e06-049.png
 86 | e06-053.png
 87 | e06-070.png
 88 | f04-032.png
 89 | f04-035.png
 90 | f04-039.png
 91 | f04-043.png
 92 | f04-053.png
 93 | f04-057.png
 94 | f04-061.png
 95 | f04-064.png
 96 | f04-068.png
 97 | f04-071.png
 98 | f04-074.png
 99 | f04-079.png
100 | f04-083.png
101 | f04-087.png
102 | f07-000.png
103 | f07-000b.png
104 | f07-002.png
105 | f07-006.png
106 | f07-009.png
107 | f07-013.png
108 | f07-016.png
109 | f07-019a.png
110 | f07-019b.png
111 | f07-021a.png
112 | f07-021b.png
113 | f07-024a.png
114 | f07-024b.png
115 | f07-028a.png
116 | f07-032a.png
117 | f07-039a.png
118 | f07-042a.png
119 | f07-046a.png
120 | f07-069.png
121 | f07-073.png
122 | f07-076a.png
123 | f07-081a.png
124 | f07-081b.png
125 | f07-084a.png
126 | f07-084b.png
127 | f07-088a.png
128 | f07-088b.png
129 | f07-092a.png
130 | f07-092b.png
131 | f07-096.png
132 | f07-101b.png
133 | g01-004.png
134 | g01-008.png
135 | g01-012.png
136 | g01-016.png
137 | g01-019.png
138 | g01-025.png
139 | g01-027.png
140 | g01-031.png
141 | g01-034.png
142 | g01-037.png
143 | g01-039.png
144 | g01-043.png
145 | g01-045.png
146 | g01-067.png
147 | g01-070.png
148 | g01-074.png
149 | g01-088.png
150 | g02-059.png
151 | g02-062.png
152 | g02-065.png
153 | g02-073.png
154 | g03-000.png
155 | g03-004.png
156 | g03-032.png
157 | g03-040.png
158 | g03-043.png
159 | g03-049.png
160 | g03-052.png
161 | g03-064.png
162 | g04-022.png
163 | g04-043.png
164 | g04-048.png
165 | g04-052.png
166 | g04-068.png
167 | g04-072.png
168 | g07-000b.png
169 | g07-079a.png
170 | m01-049.png
171 | m01-060.png
172 | m01-079.png
173 | m01-084.png
174 | m01-095.png
175 | m01-104.png
176 | m01-110.png
177 | m01-121.png
178 | m01-125.png
179 | m01-131.png
180 | m01-136.png
181 | m01-149.png
182 | m02-048.png
183 | m02-055.png
184 | m02-059.png
185 | m02-066.png
186 | m02-069.png
187 | m02-072.png
188 | m02-080.png
189 | m02-087.png
190 | m02-095.png
191 | m02-109.png
192 | m02-112.png
193 | m03-006.png
194 | m03-013.png
195 | m03-020.png
196 | m03-033.png
197 | m03-110.png
198 | m03-114.png
199 | m03-118.png
200 | m04-030.png
201 | m04-038.png
202 | m04-043.png
203 | m04-061.png
204 | m04-072.png
205 | m04-078.png
206 | m04-081.png
207 | m04-093.png
208 | m04-100.png
209 | m04-107.png
210 | m04-113.png
211 | m04-123.png
212 | m04-131.png
213 | m04-138.png
214 | m04-231.png
215 | m04-238.png
216 | m04-246.png
217 | m06-031.png
218 | m06-042.png
219 | m06-048.png
220 | m06-056.png
221 | m06-067.png
222 | m06-076.png
223 | m06-083.png
224 | m06-091.png
225 | m06-098.png
226 | m06-106.png
227 | n01-000.png
228 | n01-004.png
229 | n01-009.png
230 | n01-020.png
231 | n01-031.png
232 | n01-036.png
233 | n01-045.png
234 | n01-052.png
235 | n01-057.png
236 | n02-004.png
237 | n02-009.png
238 | n02-028.png
239 | n02-033.png
240 | n02-037.png
241 | n02-040.png
242 | n02-045.png
243 | n02-049.png
244 | n02-054.png
245 | n02-062.png
246 | n02-104.png
247 | n02-109.png
248 | n02-114.png
249 | n02-120.png
250 | n02-127.png
251 | n02-151.png
252 | n02-154.png
253 | n02-157.png
254 | n03-038.png
255 | n03-064.png
256 | n03-066.png
257 | n03-097.png
258 | n03-103.png
259 | n03-106.png
260 | n03-113.png
261 | n03-120.png
262 | n03-126.png
263 | n04-000.png
264 | n04-009.png
265 | n04-048.png
266 | n04-052.png
267 | n04-060.png
268 | n04-068.png
269 | n04-075.png
270 | n04-084.png
271 | n04-092.png
272 | n04-100.png
273 | n04-107.png
274 | n04-114.png
275 | n04-130.png
276 | n04-139.png
277 | n04-149.png
278 | n04-156.png
279 | n04-163.png
280 | n04-202.png
281 | n04-209.png
282 | n04-213.png
283 | n04-218.png
284 | n06-074.png
285 | n06-082.png
286 | n06-092.png
287 | n06-100.png
288 | n06-111.png
289 | n06-119.png
290 | n06-123.png
291 | n06-128.png
292 | n06-133.png
293 | n06-140.png
294 | p01-147.png
295 | p01-155.png
296 | p01-168.png
297 | p01-174.png
298 | p02-000.png
299 | p02-008.png
300 | p02-022.png
301 | p02-027.png
302 | p02-069.png
303 | p02-076.png
304 | p02-081.png
305 | p02-090.png
306 | p02-101.png
307 | p02-105.png
308 | p02-109.png
309 | p02-115.png
310 | p02-135.png
311 | p02-139.png
312 | p02-144.png
313 | p02-150.png
314 | p03-012.png
315 | p03-023.png
316 | p03-027.png
317 | p03-029.png
318 | p03-033.png
319 | p03-040.png
320 | p03-047.png
321 | p03-072.png
322 | p03-080.png
323 | p03-163.png
324 | p03-173.png
325 | p03-181.png
326 | p03-189.png
327 | p06-030.png
328 | p06-042.png
329 | p06-047.png
330 | p06-052.png
331 | p06-058.png
332 | p06-069.png
333 | p06-088.png
334 | p06-096.png
335 | p06-104.png
336 | r03-053.png
337 | 


--------------------------------------------------------------------------------
/iam/test.ln:
--------------------------------------------------------------------------------
   1 | g01-016-06.png
   2 | n06-100-03.png
   3 | d01-118-04.png
   4 | d04-032-08.png
   5 | n02-120-06.png
   6 | n04-060-01.png
   7 | d05-025-03.png
   8 | d06-063-07.png
   9 | n02-045-02.png
  10 | p02-150-01.png
  11 | e06-003-03.png
  12 | e01-055-06.png
  13 | m02-069-06.png
  14 | g02-065-03.png
  15 | f07-084a-08.png
  16 | d06-063-05.png
  17 | n04-100-00.png
  18 | d04-012-08.png
  19 | d07-082-05.png
  20 | n03-064-04.png
  21 | m04-123-05.png
  22 | f04-039-04.png
  23 | e06-003-09.png
  24 | d04-012-09.png
  25 | g01-031-06.png
  26 | f07-024b-05.png
  27 | e06-070-01.png
  28 | n03-066-03.png
  29 | d07-100-03.png
  30 | m01-095-07.png
  31 | f07-024b-07.png
  32 | m04-061-03.png
  33 | p06-088-02.png
  34 | c04-110-01.png
  35 | f07-024a-04.png
  36 | g03-004-01.png
  37 | m04-107-02.png
  38 | f04-071-04.png
  39 | d04-037-02.png
  40 | p06-104-01.png
  41 | c04-139-06.png
  42 | r03-053-04.png
  43 | d01-049-07.png
  44 | g01-025-03.png
  45 | r03-053-05.png
  46 | f07-088a-07.png
  47 | n02-049-04.png
  48 | f07-000b-04.png
  49 | n01-052-11.png
  50 | m06-098-00.png
  51 | n02-037-00.png
  52 | m04-113-10.png
  53 | p03-029-00.png
  54 | g01-043-09.png
  55 | c04-144-01.png
  56 | e06-046-03.png
  57 | p02-144-07.png
  58 | m03-118-04.png
  59 | f04-032-05.png
  60 | c04-150-04.png
  61 | d01-052-05.png
  62 | p06-047-06.png
  63 | f07-081a-05.png
  64 | d06-060-03.png
  65 | c06-011-08.png
  66 | c04-150-05.png
  67 | m01-104-07.png
  68 | p02-069-04.png
  69 | g01-025-01.png
  70 | p03-080-01.png
  71 | f07-002-06.png
  72 | p03-040-07.png
  73 | d04-111-09.png
  74 | d05-025-07.png
  75 | e06-070-02.png
  76 | p06-088-06.png
  77 | d06-067-01.png
  78 | d07-085-00.png
  79 | g01-070-07.png
  80 | g01-070-00.png
  81 | n04-202-01.png
  82 | f07-088a-04.png
  83 | n03-113-03.png
  84 | n01-036-02.png
  85 | n04-202-03.png
  86 | f07-092a-07.png
  87 | n06-074-06.png
  88 | c04-144-02.png
  89 | d07-093-00.png
  90 | m02-059-03.png
  91 | n04-075-02.png
  92 | d06-111-02.png
  93 | p03-023-02.png
  94 | d06-003-03.png
  95 | f07-069-04.png
  96 | m03-006-04.png
  97 | p03-040-04.png
  98 | g02-065-02.png
  99 | f07-013-09.png
 100 | m06-042-01.png
 101 | g03-064-03.png
 102 | m03-118-00.png
 103 | d04-053-00.png
 104 | m06-031-04.png
 105 | d07-089-06.png
 106 | d07-093-04.png
 107 | g01-067-04.png
 108 | p02-109-00.png
 109 | g02-065-06.png
 110 | n03-113-07.png
 111 | n06-123-08.png
 112 | p02-076-05.png
 113 | g03-064-01.png
 114 | e06-015-03.png
 115 | m01-136-06.png
 116 | m02-095-03.png
 117 | d07-082-02.png
 118 | n01-020-07.png
 119 | d06-025-04.png
 120 | d07-085-01.png
 121 | n02-151-10.png
 122 | n02-151-00.png
 123 | p02-144-05.png
 124 | m04-093-00.png
 125 | r03-053-06.png
 126 | m06-083-02.png
 127 | n04-156-04.png
 128 | f07-019b-07.png
 129 | d06-111-00.png
 130 | p02-135-07.png
 131 | n03-106-03.png
 132 | f07-016-05.png
 133 | d03-117-05.png
 134 | d06-060-05.png
 135 | m04-100-05.png
 136 | m04-078-01.png
 137 | g01-043-08.png
 138 | n02-062-00.png
 139 | n01-009-06.png
 140 | n03-064-03.png
 141 | f07-009-08.png
 142 | m03-118-05.png
 143 | e06-046-05.png
 144 | n04-114-01.png
 145 | c04-165-01.png
 146 | g01-027-09.png
 147 | g01-074-07.png
 148 | n04-213-05.png
 149 | n06-100-05.png
 150 | m04-072-06.png
 151 | n01-052-01.png
 152 | n03-120-00.png
 153 | d04-111-07.png
 154 | p03-047-05.png
 155 | m01-049-07.png
 156 | p03-029-09.png
 157 | m02-095-08.png
 158 | d01-098-06.png
 159 | n02-033-09.png
 160 | n06-074-05.png
 161 | p02-115-04.png
 162 | d06-096-02.png
 163 | m01-095-04.png
 164 | d06-041-03.png
 165 | d06-096-06.png
 166 | m04-038-04.png
 167 | n03-064-02.png
 168 | c04-165-04.png
 169 | m04-038-00.png
 170 | m06-076-03.png
 171 | d07-096-04.png
 172 | g01-004-03.png
 173 | m02-095-01.png
 174 | d05-025-08.png
 175 | m04-081-00.png
 176 | f07-009-09.png
 177 | n04-060-00.png
 178 | n04-107-03.png
 179 | g02-062-03.png
 180 | m02-109-02.png
 181 | n03-120-04.png
 182 | c06-011-04.png
 183 | n03-106-07.png
 184 | f04-061-07.png
 185 | m03-110-01.png
 186 | d06-011-08.png
 187 | n01-057-02.png
 188 | p03-189-04.png
 189 | e06-003-08.png
 190 | n04-202-02.png
 191 | f07-024a-07.png
 192 | g07-000b-02.png
 193 | f04-053-06.png
 194 | p01-174-03.png
 195 | p01-155-04.png
 196 | f04-083-08.png
 197 | m03-020-00.png
 198 | n04-048-07.png
 199 | m04-043-03.png
 200 | p02-101-05.png
 201 | f07-000b-05.png
 202 | n04-163-04.png
 203 | n01-052-08.png
 204 | g01-008-08.png
 205 | d04-050-01.png
 206 | d06-046-08.png
 207 | n04-202-06.png
 208 | d04-089-05.png
 209 | f07-081b-06.png
 210 | n04-107-04.png
 211 | g03-004-03.png
 212 | n01-020-09.png
 213 | p02-090-02.png
 214 | f04-068-04.png
 215 | d06-104-04.png
 216 | d06-113-01.png
 217 | f07-021b-02.png
 218 | d04-111-06.png
 219 | g01-004-01.png
 220 | f04-071-05.png
 221 | f07-028a-09.png
 222 | f07-046a-03.png
 223 | g04-022-05.png
 224 | n02-114-08.png
 225 | m04-061-01.png
 226 | g01-016-04.png
 227 | g03-049-06.png
 228 | m04-081-09.png
 229 | d01-123-01.png
 230 | f07-069-06.png
 231 | g01-045-05.png
 232 | n04-139-07.png
 233 | m06-083-06.png
 234 | p03-072-05.png
 235 | f04-039-00.png
 236 | m02-069-01.png
 237 | e06-053-03.png
 238 | g03-052-05.png
 239 | g04-022-04.png
 240 | n04-052-03.png
 241 | f04-064-01.png
 242 | p03-023-01.png
 243 | n01-045-04.png
 244 | g04-072-07.png
 245 | d04-111-05.png
 246 | m02-112-05.png
 247 | f04-053-01.png
 248 | d06-046-06.png
 249 | f07-016-07.png
 250 | d07-100-04.png
 251 | d06-008-09.png
 252 | p03-047-07.png
 253 | f07-016-04.png
 254 | n06-100-02.png
 255 | d04-066-06.png
 256 | n02-028-00.png
 257 | g02-059-01.png
 258 | p06-088-00.png
 259 | d01-098-03.png
 260 | d06-111-04.png
 261 | m02-072-01.png
 262 | n04-052-04.png
 263 | d06-046-03.png
 264 | p03-047-01.png
 265 | e06-026-01.png
 266 | f07-002-00.png
 267 | d04-125-00.png
 268 | e06-030-04.png
 269 | f04-079-03.png
 270 | g01-037-02.png
 271 | n02-054-09.png
 272 | p02-090-03.png
 273 | m04-043-01.png
 274 | m04-081-01.png
 275 | f07-069-05.png
 276 | d05-040-05.png
 277 | d04-016-06.png
 278 | r03-053-01.png
 279 | n06-140-06.png
 280 | p06-030-02.png
 281 | d05-025-06.png
 282 | e06-049-03.png
 283 | n06-082-04.png
 284 | p02-150-00.png
 285 | p03-047-02.png
 286 | d01-080-00.png
 287 | n03-103-09.png
 288 | n01-031-08.png
 289 | g03-052-01.png
 290 | n02-109-01.png
 291 | e01-055-03.png
 292 | m01-084-02.png
 293 | p03-047-04.png
 294 | d06-086-02.png
 295 | n02-104-07.png
 296 | m04-131-02.png
 297 | p02-076-10.png
 298 | c04-150-07.png
 299 | g04-048-04.png
 300 | f07-000-01.png
 301 | p06-069-03.png
 302 | f07-024b-09.png
 303 | n04-009-03.png
 304 | n01-045-09.png
 305 | n01-020-06.png
 306 | n02-028-01.png
 307 | p03-072-00.png
 308 | f07-024a-01.png
 309 | n06-092-03.png
 310 | p03-181-05.png
 311 | d07-089-07.png
 312 | m04-238-00.png
 313 | d05-025-02.png
 314 | n03-106-06.png
 315 | p02-105-01.png
 316 | m01-060-06.png
 317 | g03-043-04.png
 318 | f07-088a-03.png
 319 | m02-059-06.png
 320 | d06-020-02.png
 321 | p02-008-04.png
 322 | n04-139-08.png
 323 | f04-068-00.png
 324 | m04-123-09.png
 325 | d07-096-02.png
 326 | g04-022-07.png
 327 | m01-110-07.png
 328 | n04-149-02.png
 329 | g01-008-07.png
 330 | d07-082-03.png
 331 | d01-049-01.png
 332 | f07-042a-00.png
 333 | d01-080-03.png
 334 | d07-100-02.png
 335 | p03-047-06.png
 336 | n06-140-00.png
 337 | n04-100-07.png
 338 | n04-107-01.png
 339 | p02-000-06.png
 340 | f07-073-08.png
 341 | n01-009-00.png
 342 | n02-104-05.png
 343 | d05-040-00.png
 344 | g04-043-01.png
 345 | n02-037-06.png
 346 | p03-181-06.png
 347 | d06-072-05.png
 348 | p03-163-01.png
 349 | f04-061-06.png
 350 | n02-127-07.png
 351 | n04-107-06.png
 352 | d01-019-02.png
 353 | f07-039a-03.png
 354 | n04-084-03.png
 355 | n02-114-06.png
 356 | f07-006-00.png
 357 | d04-062-03.png
 358 | f07-019a-00.png
 359 | d04-016-00.png
 360 | g01-025-02.png
 361 | d05-040-03.png
 362 | d05-025-09.png
 363 | p06-104-04.png
 364 | m04-100-02.png
 365 | m06-042-02.png
 366 | e01-055-04.png
 367 | g01-034-01.png
 368 | p06-030-06.png
 369 | n01-045-02.png
 370 | m04-061-04.png
 371 | p03-033-04.png
 372 | f04-074-07.png
 373 | p01-147-00.png
 374 | f07-032a-08.png
 375 | c04-150-00.png
 376 | m04-131-07.png
 377 | n02-054-08.png
 378 | c04-165-02.png
 379 | f04-087-00.png
 380 | p06-042-05.png
 381 | e06-015-02.png
 382 | n01-052-07.png
 383 | g03-040-01.png
 384 | d06-020-09.png
 385 | m02-069-08.png
 386 | n04-156-00.png
 387 | e06-046-01.png
 388 | n02-054-01.png
 389 | n02-154-02.png
 390 | c04-139-00.png
 391 | d06-011-02.png
 392 | d06-037-02.png
 393 | g01-037-03.png
 394 | n02-004-06.png
 395 | d04-086-03.png
 396 | m02-080-04.png
 397 | d06-113-00.png
 398 | e06-021-09.png
 399 | m04-238-05.png
 400 | m02-087-00.png
 401 | f04-057-05.png
 402 | d04-050-00.png
 403 | f07-088b-04.png
 404 | m01-095-05.png
 405 | n02-040-01.png
 406 | c04-139-02.png
 407 | m02-048-10.png
 408 | g04-043-07.png
 409 | n06-133-06.png
 410 | f07-024a-06.png
 411 | g03-040-04.png
 412 | m04-081-04.png
 413 | f07-032a-07.png
 414 | g01-037-00.png
 415 | m01-079-00.png
 416 | d01-052-01.png
 417 | p06-047-00.png
 418 | n01-057-03.png
 419 | n01-031-05.png
 420 | d06-008-05.png
 421 | d04-125-03.png
 422 | n01-020-05.png
 423 | d06-107-06.png
 424 | d06-086-01.png
 425 | m01-104-05.png
 426 | d07-089-10.png
 427 | d05-025-01.png
 428 | f07-073-07.png
 429 | m03-114-01.png
 430 | m04-100-06.png
 431 | p02-081-07.png
 432 | d01-098-00.png
 433 | f04-083-01.png
 434 | n06-133-04.png
 435 | d04-066-07.png
 436 | g02-062-04.png
 437 | g03-032-05.png
 438 | n02-054-07.png
 439 | d01-049-05.png
 440 | p01-147-11.png
 441 | n01-009-03.png
 442 | m04-138-03.png
 443 | d01-085-01.png
 444 | m06-031-01.png
 445 | e06-033-07.png
 446 | m01-110-08.png
 447 | d06-046-09.png
 448 | p03-173-05.png
 449 | f04-068-01.png
 450 | n02-033-08.png
 451 | m04-131-03.png
 452 | p01-147-09.png
 453 | n02-049-00.png
 454 | p03-012-07.png
 455 | f07-024a-03.png
 456 | p06-069-09.png
 457 | p06-069-02.png
 458 | g04-072-01.png
 459 | p02-105-00.png
 460 | d06-100-01.png
 461 | m02-055-01.png
 462 | g01-045-00.png
 463 | n03-097-01.png
 464 | d01-085-00.png
 465 | m03-033-04.png
 466 | m06-031-03.png
 467 | f04-074-02.png
 468 | d06-067-05.png
 469 | d01-098-04.png
 470 | n02-054-06.png
 471 | n04-052-05.png
 472 | d06-100-03.png
 473 | m01-136-04.png
 474 | p02-135-02.png
 475 | p03-173-08.png
 476 | p02-090-07.png
 477 | m01-049-04.png
 478 | m02-112-03.png
 479 | n04-092-07.png
 480 | g01-039-00.png
 481 | m01-079-04.png
 482 | d04-028-04.png
 483 | n04-114-00.png
 484 | f07-084a-09.png
 485 | f04-064-08.png
 486 | e06-053-05.png
 487 | g04-068-03.png
 488 | g04-072-00.png
 489 | f07-073-02.png
 490 | m04-107-06.png
 491 | m06-083-00.png
 492 | p02-144-03.png
 493 | m01-136-07.png
 494 | n06-100-01.png
 495 | g02-059-05.png
 496 | m04-138-09.png
 497 | f07-016-02.png
 498 | d06-107-04.png
 499 | g01-039-05.png
 500 | d01-085-02.png
 501 | m01-049-05.png
 502 | p02-135-00.png
 503 | p03-173-01.png
 504 | f07-021b-05.png
 505 | e01-055-00.png
 506 | p06-047-04.png
 507 | f07-021b-06.png
 508 | g03-052-06.png
 509 | f04-064-05.png
 510 | m04-038-02.png
 511 | f04-035-05.png
 512 | g01-031-05.png
 513 | m04-123-08.png
 514 | g01-012-06.png
 515 | m04-081-10.png
 516 | m02-055-04.png
 517 | d06-072-02.png
 518 | d06-041-01.png
 519 | n02-109-05.png
 520 | p02-000-00.png
 521 | g01-039-02.png
 522 | f07-084a-10.png
 523 | e06-021-10.png
 524 | g01-025-05.png
 525 | n02-004-05.png
 526 | f07-046a-05.png
 527 | f07-046a-01.png
 528 | p03-173-03.png
 529 | n06-092-01.png
 530 | n03-038-06.png
 531 | g04-052-01.png
 532 | g01-016-05.png
 533 | d01-123-03.png
 534 | f07-046a-02.png
 535 | n06-123-02.png
 536 | d06-072-04.png
 537 | g04-022-03.png
 538 | m02-048-08.png
 539 | f04-087-02.png
 540 | d01-098-01.png
 541 | f07-092a-06.png
 542 | e06-049-05.png
 543 | n02-054-03.png
 544 | m03-114-08.png
 545 | n04-060-06.png
 546 | f04-087-05.png
 547 | n02-157-04.png
 548 | d06-027-05.png
 549 | p03-173-07.png
 550 | p06-069-07.png
 551 | n06-128-05.png
 552 | m06-098-06.png
 553 | n06-092-07.png
 554 | d06-027-03.png
 555 | n06-140-07.png
 556 | g03-064-00.png
 557 | g07-000b-05.png
 558 | m04-123-02.png
 559 | m06-031-00.png
 560 | n01-004-00.png
 561 | f04-039-06.png
 562 | p03-080-02.png
 563 | e06-015-01.png
 564 | g01-031-01.png
 565 | m04-100-01.png
 566 | n01-045-01.png
 567 | p02-109-06.png
 568 | m04-100-00.png
 569 | p02-150-02.png
 570 | g02-073-05.png
 571 | f07-013-07.png
 572 | m06-106-05.png
 573 | f07-084a-05.png
 574 | d04-089-01.png
 575 | f07-076a-09.png
 576 | f07-028a-07.png
 577 | n04-009-00.png
 578 | d04-032-07.png
 579 | d07-100-06.png
 580 | d07-096-05.png
 581 | p03-163-00.png
 582 | p06-104-03.png
 583 | g03-049-07.png
 584 | p02-101-07.png
 585 | p06-104-05.png
 586 | n01-020-02.png
 587 | n02-033-01.png
 588 | g01-088-01.png
 589 | m02-069-04.png
 590 | m02-109-07.png
 591 | d04-053-05.png
 592 | m06-106-07.png
 593 | n03-113-08.png
 594 | d06-067-09.png
 595 | d01-123-06.png
 596 | n02-127-02.png
 597 | d06-107-07.png
 598 | d06-060-07.png
 599 | e06-026-09.png
 600 | m02-059-02.png
 601 | p01-155-06.png
 602 | n02-157-01.png
 603 | m01-060-00.png
 604 | d04-062-02.png
 605 | d06-008-07.png
 606 | n04-218-00.png
 607 | g01-067-00.png
 608 | n02-033-05.png
 609 | e06-030-06.png
 610 | m01-149-05.png
 611 | g04-048-07.png
 612 | f04-043-06.png
 613 | d04-012-07.png
 614 | d04-032-04.png
 615 | f07-081b-03.png
 616 | p01-174-07.png
 617 | p02-000-01.png
 618 | g01-019-01.png
 619 | e06-053-07.png
 620 | g02-073-03.png
 621 | r03-053-08.png
 622 | m06-048-01.png
 623 | n02-120-01.png
 624 | n01-000-03.png
 625 | c04-134-03.png
 626 | p03-033-03.png
 627 | f04-074-04.png
 628 | p06-058-01.png
 629 | d06-100-07.png
 630 | p06-096-03.png
 631 | p06-096-00.png
 632 | n01-036-01.png
 633 | d06-011-00.png
 634 | p03-023-05.png
 635 | r03-053-07.png
 636 | d06-011-06.png
 637 | d04-131-03.png
 638 | m02-069-05.png
 639 | m01-149-07.png
 640 | g01-025-08.png
 641 | e06-015-08.png
 642 | d05-025-05.png
 643 | g04-022-00.png
 644 | d06-025-01.png
 645 | g01-012-03.png
 646 | g02-059-00.png
 647 | n01-057-04.png
 648 | p02-101-04.png
 649 | n02-127-04.png
 650 | n04-218-08.png
 651 | m06-076-06.png
 652 | g03-040-00.png
 653 | m04-138-04.png
 654 | d06-107-03.png
 655 | p02-008-02.png
 656 | d07-082-01.png
 657 | e06-026-05.png
 658 | p03-072-02.png
 659 | m04-061-00.png
 660 | n06-128-08.png
 661 | p02-101-00.png
 662 | m01-131-05.png
 663 | m04-123-00.png
 664 | d04-016-04.png
 665 | g01-074-03.png
 666 | m02-109-03.png
 667 | d01-123-00.png
 668 | p06-047-02.png
 669 | c06-011-01.png
 670 | m01-121-04.png
 671 | m01-110-01.png
 672 | p03-173-02.png
 673 | f07-013-05.png
 674 | n06-133-01.png
 675 | p01-155-08.png
 676 | m02-059-00.png
 677 | m04-093-01.png
 678 | n04-156-02.png
 679 | n04-048-03.png
 680 | p02-105-04.png
 681 | m03-020-01.png
 682 | c04-110-03.png
 683 | d01-049-06.png
 684 | n01-020-01.png
 685 | g02-059-07.png
 686 | n04-068-05.png
 687 | e06-033-08.png
 688 | d06-086-10.png
 689 | e06-000-07.png
 690 | p06-042-03.png
 691 | f07-032a-10.png
 692 | f07-088a-01.png
 693 | n01-000-04.png
 694 | d06-037-03.png
 695 | m02-080-09.png
 696 | m06-083-03.png
 697 | e06-049-06.png
 698 | g03-032-07.png
 699 | d03-117-01.png
 700 | n04-202-00.png
 701 | g01-039-03.png
 702 | d04-047-01.png
 703 | n01-000-02.png
 704 | m06-076-04.png
 705 | f07-000b-06.png
 706 | e06-070-03.png
 707 | n03-064-00.png
 708 | m04-093-08.png
 709 | m03-006-06.png
 710 | d06-100-00.png
 711 | p06-058-09.png
 712 | m06-076-00.png
 713 | m04-030-02.png
 714 | d06-104-09.png
 715 | m04-246-01.png
 716 | p02-101-01.png
 717 | g07-000b-00.png
 718 | m04-123-04.png
 719 | n06-128-01.png
 720 | f04-043-02.png
 721 | d01-104-04.png
 722 | d04-050-02.png
 723 | g03-040-02.png
 724 | m01-149-01.png
 725 | m04-061-02.png
 726 | f04-087-04.png
 727 | p03-027-00.png
 728 | n02-045-04.png
 729 | d04-125-05.png
 730 | f07-088b-02.png
 731 | n01-036-00.png
 732 | f07-076a-01.png
 733 | g04-048-06.png
 734 | n03-126-00.png
 735 | m01-121-02.png
 736 | p02-135-01.png
 737 | g03-064-07.png
 738 | m04-131-04.png
 739 | d06-037-04.png
 740 | d04-089-02.png
 741 | m04-072-05.png
 742 | d04-131-01.png
 743 | d01-085-04.png
 744 | d07-102-01.png
 745 | e06-053-00.png
 746 | g01-008-05.png
 747 | e06-010-05.png
 748 | d06-060-08.png
 749 | d05-013-07.png
 750 | d05-008-03.png
 751 | d07-096-07.png
 752 | n02-028-09.png
 753 | d04-021-04.png
 754 | d06-037-05.png
 755 | e06-010-06.png
 756 | d06-086-07.png
 757 | f07-021a-05.png
 758 | d04-101-02.png
 759 | p06-088-07.png
 760 | m02-095-06.png
 761 | p03-072-04.png
 762 | g07-000b-01.png
 763 | d04-089-04.png
 764 | f07-088b-06.png
 765 | d04-050-04.png
 766 | f07-002-03.png
 767 | d04-066-09.png
 768 | m04-113-02.png
 769 | m04-246-04.png
 770 | p03-181-02.png
 771 | d04-111-08.png
 772 | p03-029-07.png
 773 | e06-021-05.png
 774 | f07-019a-01.png
 775 | f07-009-04.png
 776 | g01-027-04.png
 777 | d06-020-06.png
 778 | e06-010-04.png
 779 | n02-114-09.png
 780 | m06-042-05.png
 781 | n01-020-03.png
 782 | f07-076a-04.png
 783 | n02-157-05.png
 784 | f04-079-09.png
 785 | m01-131-06.png
 786 | d04-037-00.png
 787 | m01-149-03.png
 788 | m04-113-00.png
 789 | d04-016-02.png
 790 | n02-127-05.png
 791 | d06-041-02.png
 792 | n01-057-01.png
 793 | g03-052-00.png
 794 | p02-069-07.png
 795 | d06-096-01.png
 796 | f07-019b-01.png
 797 | c04-134-00.png
 798 | m04-072-04.png
 799 | f04-057-07.png
 800 | n04-107-07.png
 801 | g03-000-01.png
 802 | n04-139-01.png
 803 | g01-016-07.png
 804 | n03-120-05.png
 805 | m02-055-00.png
 806 | m02-080-07.png
 807 | d05-030-06.png
 808 | f07-101b-03.png
 809 | d01-123-07.png
 810 | m03-110-03.png
 811 | n03-038-01.png
 812 | n04-130-05.png
 813 | p01-174-05.png
 814 | d04-032-02.png
 815 | n02-154-05.png
 816 | d04-053-02.png
 817 | g01-067-09.png
 818 | m04-131-00.png
 819 | g03-032-04.png
 820 | f07-000-06.png
 821 | m03-033-05.png
 822 | n02-009-04.png
 823 | p02-076-04.png
 824 | p02-144-00.png
 825 | m01-125-09.png
 826 | n02-154-03.png
 827 | f04-043-05.png
 828 | n03-103-03.png
 829 | p03-012-00.png
 830 | p06-052-00.png
 831 | m04-081-08.png
 832 | n04-149-01.png
 833 | m06-067-05.png
 834 | d06-063-04.png
 835 | d04-012-04.png
 836 | n01-009-02.png
 837 | f07-101b-00.png
 838 | n04-218-04.png
 839 | p06-069-01.png
 840 | d04-086-04.png
 841 | p06-030-04.png
 842 | d07-093-07.png
 843 | n02-104-01.png
 844 | g03-043-06.png
 845 | d06-056-06.png
 846 | m01-125-02.png
 847 | p03-029-01.png
 848 | g01-088-08.png
 849 | n02-157-03.png
 850 | p01-147-02.png
 851 | f07-024b-01.png
 852 | p02-076-01.png
 853 | f04-068-02.png
 854 | m03-006-00.png
 855 | m04-113-09.png
 856 | n04-209-02.png
 857 | p01-147-05.png
 858 | m04-030-03.png
 859 | e06-030-07.png
 860 | d06-104-06.png
 861 | d04-047-04.png
 862 | d04-096-07.png
 863 | f07-046a-07.png
 864 | g03-043-02.png
 865 | g01-043-01.png
 866 | n04-052-02.png
 867 | d01-085-05.png
 868 | m02-055-08.png
 869 | p06-047-03.png
 870 | p03-181-04.png
 871 | m02-069-00.png
 872 | d06-111-03.png
 873 | d04-047-05.png
 874 | p03-072-07.png
 875 | c04-139-03.png
 876 | p02-139-05.png
 877 | g01-037-08.png
 878 | n03-103-05.png
 879 | m03-006-05.png
 880 | n02-045-03.png
 881 | p03-033-02.png
 882 | g01-045-02.png
 883 | m02-066-05.png
 884 | m03-013-05.png
 885 | g04-052-05.png
 886 | f07-013-01.png
 887 | n06-128-07.png
 888 | n02-114-03.png
 889 | g03-040-03.png
 890 | n04-075-01.png
 891 | p03-047-08.png
 892 | g01-067-08.png
 893 | d04-086-06.png
 894 | n02-004-08.png
 895 | m02-087-06.png
 896 | p02-008-01.png
 897 | d04-058-03.png
 898 | f07-081a-02.png
 899 | e06-021-03.png
 900 | d04-086-02.png
 901 | g04-072-04.png
 902 | g01-012-04.png
 903 | f04-057-03.png
 904 | n04-084-05.png
 905 | n06-119-03.png
 906 | m03-110-09.png
 907 | n06-119-05.png
 908 | m06-076-05.png
 909 | n03-113-00.png
 910 | d01-019-05.png
 911 | e06-033-04.png
 912 | n04-092-02.png
 913 | m04-231-04.png
 914 | d06-027-02.png
 915 | n03-120-09.png
 916 | g01-004-05.png
 917 | d01-104-05.png
 918 | e06-053-08.png
 919 | f07-084a-03.png
 920 | m01-136-03.png
 921 | n02-028-06.png
 922 | p06-096-07.png
 923 | c04-144-04.png
 924 | m02-072-05.png
 925 | f07-009-01.png
 926 | f07-096-01.png
 927 | g01-074-09.png
 928 | m06-106-02.png
 929 | f07-092b-03.png
 930 | f07-024b-04.png
 931 | n04-156-01.png
 932 | n02-120-02.png
 933 | f04-061-03.png
 934 | g01-031-03.png
 935 | e06-070-00.png
 936 | d04-086-00.png
 937 | d04-131-00.png
 938 | n02-151-02.png
 939 | f07-096-05.png
 940 | c04-165-05.png
 941 | f07-002-05.png
 942 | n06-074-01.png
 943 | d03-117-06.png
 944 | n04-156-05.png
 945 | n04-213-03.png
 946 | f04-053-04.png
 947 | n04-213-07.png
 948 | g01-008-04.png
 949 | m01-084-03.png
 950 | c04-170-02.png
 951 | g03-052-04.png
 952 | g01-074-04.png
 953 | g02-065-04.png
 954 | g03-032-01.png
 955 | d01-052-07.png
 956 | f04-068-06.png
 957 | g01-043-07.png
 958 | n01-031-06.png
 959 | g07-079a-08.png
 960 | p02-027-04.png
 961 | f07-092b-05.png
 962 | g01-045-09.png
 963 | n04-139-00.png
 964 | g01-027-07.png
 965 | n02-040-02.png
 966 | p06-058-05.png
 967 | e06-049-08.png
 968 | p03-012-03.png
 969 | r03-053-02.png
 970 | g02-062-02.png
 971 | f07-081a-09.png
 972 | m02-109-06.png
 973 | d04-053-03.png
 974 | m04-231-03.png
 975 | p02-101-03.png
 976 | c04-170-00.png
 977 | m04-238-04.png
 978 | e06-010-01.png
 979 | g04-068-04.png
 980 | p06-052-02.png
 981 | n03-106-08.png
 982 | g01-031-09.png
 983 | f07-046a-08.png
 984 | f07-039a-06.png
 985 | n04-068-04.png
 986 | m01-060-01.png
 987 | m04-123-01.png
 988 | g01-037-01.png
 989 | f04-039-07.png
 990 | d05-021-06.png
 991 | d01-118-03.png
 992 | n02-114-01.png
 993 | d03-117-02.png
 994 | e06-003-07.png
 995 | f07-009-07.png
 996 | n03-066-06.png
 997 | d04-037-01.png
 998 | m04-107-10.png
 999 | d06-008-01.png
1000 | g03-052-03.png
1001 | m02-066-02.png
1002 | n04-114-02.png
1003 | e06-021-07.png
1004 | n02-040-03.png
1005 | n02-151-01.png
1006 | n04-209-05.png
1007 | d06-067-00.png
1008 | g04-072-03.png
1009 | m01-136-08.png
1010 | n04-084-04.png
1011 | p03-080-05.png
1012 | n04-060-05.png
1013 | p06-058-02.png
1014 | m06-067-06.png
1015 | n01-004-01.png
1016 | d06-008-08.png
1017 | d06-020-01.png
1018 | g01-070-08.png
1019 | m03-110-08.png
1020 | g07-079a-03.png
1021 | m03-006-09.png
1022 | n02-062-05.png
1023 | d06-104-02.png
1024 | d07-102-06.png
1025 | m06-083-01.png
1026 | f07-069-08.png
1027 | m04-246-00.png
1028 | d06-027-04.png
1029 | d04-032-06.png
1030 | f07-028a-08.png
1031 | m02-048-05.png
1032 | n02-037-03.png
1033 | g03-064-05.png
1034 | m04-038-03.png
1035 | n03-103-07.png
1036 | d05-030-04.png
1037 | m04-030-04.png
1038 | n03-106-10.png
1039 | p02-008-03.png
1040 | g01-039-08.png
1041 | p02-135-06.png
1042 | g01-088-02.png
1043 | n04-149-07.png
1044 | p03-012-04.png
1045 | g01-045-04.png
1046 | d06-096-05.png
1047 | f07-084a-06.png
1048 | g01-070-01.png
1049 | n03-066-04.png
1050 | n04-068-01.png
1051 | g04-052-06.png
1052 | n06-119-04.png
1053 | m04-093-02.png
1054 | e06-026-00.png
1055 | e01-055-02.png
1056 | d04-053-09.png
1057 | g04-068-01.png
1058 | m06-056-02.png
1059 | f07-032a-02.png
1060 | m04-231-02.png
1061 | p03-080-03.png
1062 | f04-043-04.png
1063 | n01-057-06.png
1064 | p06-047-10.png
1065 | d06-086-00.png
1066 | n01-052-10.png
1067 | e01-055-09.png
1068 | m04-078-05.png
1069 | e06-053-06.png
1070 | e06-000-02.png
1071 | n04-009-04.png
1072 | e06-049-04.png
1073 | m02-048-06.png
1074 | n06-111-00.png
1075 | d06-008-04.png
1076 | d06-060-06.png
1077 | p03-027-05.png
1078 | g01-034-00.png
1079 | c04-150-01.png
1080 | g04-068-08.png
1081 | n01-004-02.png
1082 | n04-052-06.png
1083 | n06-123-09.png
1084 | n06-133-07.png
1085 | n02-004-02.png
1086 | d05-013-05.png
1087 | e06-015-04.png
1088 | p02-105-02.png
1089 | p02-115-05.png
1090 | d07-100-05.png
1091 | p02-105-07.png
1092 | p03-029-08.png
1093 | n04-114-07.png
1094 | p06-069-05.png
1095 | n04-218-09.png
1096 | m01-104-04.png
1097 | d06-025-03.png
1098 | g01-074-02.png
1099 | g01-008-06.png
1100 | p01-168-06.png
1101 | g01-037-04.png
1102 | d06-063-02.png
1103 | m01-125-06.png
1104 | f07-024a-00.png
1105 | n06-111-02.png
1106 | e06-049-07.png
1107 | d05-030-08.png
1108 | d01-052-06.png
1109 | g03-043-08.png
1110 | g01-019-08.png
1111 | n01-052-12.png
1112 | n04-114-08.png
1113 | p01-168-00.png
1114 | d07-093-06.png
1115 | n03-097-00.png
1116 | d06-100-04.png
1117 | n04-218-01.png
1118 | p01-174-01.png
1119 | m06-067-01.png
1120 | n06-111-04.png
1121 | d06-096-07.png
1122 | m02-055-09.png
1123 | m02-080-00.png
1124 | e06-021-00.png
1125 | d04-021-02.png
1126 | m01-104-06.png
1127 | m01-149-06.png
1128 | n02-054-02.png
1129 | n04-213-01.png
1130 | n04-130-00.png
1131 | n02-151-09.png
1132 | p02-135-04.png
1133 | p02-105-05.png
1134 | c04-139-04.png
1135 | f04-057-00.png
1136 | c04-134-06.png
1137 | d06-056-04.png
1138 | g04-052-03.png
1139 | f04-032-03.png
1140 | f04-039-02.png
1141 | p06-030-03.png
1142 | f07-084b-03.png
1143 | p02-000-04.png
1144 | f07-021b-08.png
1145 | g02-059-04.png
1146 | n03-097-06.png
1147 | n06-100-07.png
1148 | p03-040-08.png
1149 | d06-107-08.png
1150 | g02-062-08.png
1151 | d01-098-07.png
1152 | n04-100-06.png
1153 | p03-023-03.png
1154 | n04-149-04.png
1155 | n02-157-08.png
1156 | p03-047-03.png
1157 | f07-019a-07.png
1158 | g04-072-05.png
1159 | n03-038-04.png
1160 | n04-163-02.png
1161 | m04-072-08.png
1162 | f04-074-01.png
1163 | m04-078-00.png
1164 | m03-110-04.png
1165 | p02-008-06.png
1166 | f04-035-00.png
1167 | c06-011-00.png
1168 | p03-029-06.png
1169 | m03-110-07.png
1170 | f04-079-10.png
1171 | f07-019b-05.png
1172 | e06-070-09.png
1173 | d04-131-05.png
1174 | m04-038-05.png
1175 | d04-021-07.png
1176 | p02-144-08.png
1177 | p02-081-06.png
1178 | d06-067-03.png
1179 | n02-062-07.png
1180 | m04-138-08.png
1181 | d07-082-00.png
1182 | f07-076a-06.png
1183 | e06-010-00.png
1184 | d04-111-02.png
1185 | d01-019-00.png
1186 | d07-089-02.png
1187 | p02-069-00.png
1188 | g04-043-04.png
1189 | f07-096-06.png
1190 | m01-084-01.png
1191 | m02-072-00.png
1192 | n02-049-02.png
1193 | p02-144-02.png
1194 | d04-028-05.png
1195 | f04-061-01.png
1196 | n04-060-03.png
1197 | n04-068-06.png
1198 | m04-072-00.png
1199 | m04-107-05.png
1200 | n06-119-02.png
1201 | p01-147-10.png
1202 | d06-107-00.png
1203 | d05-030-01.png
1204 | d07-102-00.png
1205 | m04-123-07.png
1206 | g01-027-03.png
1207 | d06-076-02.png
1208 | n02-004-07.png
1209 | n01-045-03.png
1210 | d04-096-01.png
1211 | p02-101-06.png
1212 | n04-000-04.png
1213 | n02-109-04.png
1214 | d04-037-03.png
1215 | m04-113-01.png
1216 | n04-052-01.png
1217 | n04-149-05.png
1218 | c04-110-02.png
1219 | n01-045-06.png
1220 | f07-032a-00.png
1221 | g01-088-04.png
1222 | g01-039-06.png
1223 | d06-056-07.png
1224 | m01-104-01.png
1225 | g01-031-10.png
1226 | m02-080-08.png
1227 | g03-032-03.png
1228 | d06-020-07.png
1229 | f07-081a-06.png
1230 | e06-003-02.png
1231 | m04-138-01.png
1232 | d06-104-03.png
1233 | m04-131-08.png
1234 | c04-116-03.png
1235 | p03-072-01.png
1236 | d06-037-07.png
1237 | m01-110-02.png
1238 | d06-104-01.png
1239 | f07-019b-09.png
1240 | g01-043-00.png
1241 | f07-088b-05.png
1242 | e06-021-02.png
1243 | d06-063-00.png
1244 | n01-052-02.png
1245 | m03-033-00.png
1246 | g01-070-05.png
1247 | m01-121-06.png
1248 | d07-089-03.png
1249 | p02-115-03.png
1250 | n02-062-01.png
1251 | n04-052-00.png
1252 | n01-009-07.png
1253 | f04-043-00.png
1254 | p06-088-08.png
1255 | m06-048-06.png
1256 | g04-043-02.png
1257 | f07-039a-05.png
1258 | n02-127-06.png
1259 | p06-069-00.png
1260 | n06-092-00.png
1261 | n04-048-09.png
1262 | e06-015-09.png
1263 | f04-053-05.png
1264 | n03-113-09.png
1265 | f04-087-03.png
1266 | p01-147-01.png
1267 | m01-125-08.png
1268 | g04-068-07.png
1269 | n04-139-03.png
1270 | g01-031-04.png
1271 | f04-032-00.png
1272 | g01-045-07.png
1273 | f07-042a-06.png
1274 | d06-076-03.png
1275 | p03-023-04.png
1276 | d04-111-11.png
1277 | m02-095-04.png
1278 | n04-009-06.png
1279 | m02-059-01.png
1280 | d01-118-06.png
1281 | m02-055-10.png
1282 | d04-066-04.png
1283 | f07-092a-05.png
1284 | n06-100-04.png
1285 | d05-008-02.png
1286 | n04-107-08.png
1287 | m04-030-00.png
1288 | g03-040-05.png
1289 | g07-079a-11.png
1290 | m01-149-00.png
1291 | n01-000-07.png
1292 | n01-057-05.png
1293 | g01-043-04.png
1294 | p02-144-06.png
1295 | g02-062-07.png
1296 | d04-032-03.png
1297 | f07-081b-01.png
1298 | d06-072-00.png
1299 | m01-136-02.png
1300 | p02-109-04.png
1301 | d01-049-03.png
1302 | n02-054-04.png
1303 | g03-049-04.png
1304 | n01-045-05.png
1305 | p06-042-00.png
1306 | d07-093-01.png
1307 | d06-056-03.png
1308 | g04-043-03.png
1309 | p02-105-06.png
1310 | f07-069-09.png
1311 | f04-061-00.png
1312 | p06-042-07.png
1313 | f07-101b-01.png
1314 | n02-157-06.png
1315 | n04-209-04.png
1316 | g02-062-05.png
1317 | n02-009-02.png
1318 | g01-045-06.png
1319 | d07-096-09.png
1320 | c04-170-03.png
1321 | f07-019a-05.png
1322 | p02-076-07.png
1323 | m01-049-11.png
1324 | f07-069-07.png
1325 | n02-040-05.png
1326 | n03-113-02.png
1327 | n04-107-05.png
1328 | p02-008-05.png
1329 | p02-000-02.png
1330 | g01-025-09.png
1331 | g07-079a-01.png
1332 | m02-080-02.png
1333 | d04-050-03.png
1334 | n04-092-03.png
1335 | n04-075-07.png
1336 | g01-008-03.png
1337 | d04-021-06.png
1338 | g02-073-00.png
1339 | m04-093-03.png
1340 | e06-021-01.png
1341 | m06-048-05.png
1342 | n04-052-07.png
1343 | n02-109-03.png
1344 | n06-123-06.png
1345 | f04-064-07.png
1346 | f04-064-00.png
1347 | g02-059-02.png
1348 | g01-088-07.png
1349 | p03-027-01.png
1350 | d04-058-00.png
1351 | n04-092-01.png
1352 | d01-016-03.png
1353 | d04-131-02.png
1354 | e06-033-03.png
1355 | p03-080-06.png
1356 | p06-052-04.png
1357 | n04-163-00.png
1358 | n06-100-09.png
1359 | f04-074-06.png
1360 | d06-046-01.png
1361 | m01-121-08.png
1362 | n04-100-04.png
1363 | n06-111-07.png
1364 | p02-081-05.png
1365 | m04-113-05.png
1366 | n06-082-01.png
1367 | f07-042a-05.png
1368 | m06-091-06.png
1369 | f07-084a-00.png
1370 | m04-138-00.png
1371 | g01-019-02.png
1372 | g01-034-08.png
1373 | g07-079a-09.png
1374 | d05-040-02.png
1375 | g01-004-00.png
1376 | d04-089-00.png
1377 | d01-104-07.png
1378 | n06-123-05.png
1379 | d04-101-00.png
1380 | g02-059-03.png
1381 | m01-125-10.png
1382 | p02-076-09.png
1383 | g07-000b-09.png
1384 | p06-058-07.png
1385 | p02-076-06.png
1386 | f07-069-01.png
1387 | g01-034-04.png
1388 | p06-052-01.png
1389 | g01-012-01.png
1390 | g02-059-06.png
1391 | g01-019-00.png
1392 | f04-079-06.png
1393 | p02-022-05.png
1394 | m06-056-05.png
1395 | n06-140-02.png
1396 | p02-069-08.png
1397 | n06-092-02.png
1398 | d06-011-01.png
1399 | m02-112-01.png
1400 | g07-079a-10.png
1401 | d05-025-00.png
1402 | m06-056-04.png
1403 | g04-072-02.png
1404 | g01-025-04.png
1405 | d05-030-02.png
1406 | n01-031-04.png
1407 | f07-021b-00.png
1408 | m06-067-00.png
1409 | g03-064-02.png
1410 | d07-093-03.png
1411 | m01-121-00.png
1412 | f07-021a-01.png
1413 | e06-015-07.png
1414 | g03-000-02.png
1415 | m06-042-03.png
1416 | d06-046-04.png
1417 | f04-083-06.png
1418 | n06-133-05.png
1419 | m04-072-09.png
1420 | d01-104-06.png
1421 | f07-096-02.png
1422 | d06-046-05.png
1423 | m04-072-03.png
1424 | n02-151-03.png
1425 | n01-057-07.png
1426 | p03-027-02.png
1427 | m04-081-05.png
1428 | p02-109-02.png
1429 | f07-039a-02.png
1430 | n04-000-03.png
1431 | n04-163-07.png
1432 | n06-092-06.png
1433 | f04-057-09.png
1434 | m03-013-03.png
1435 | d06-107-05.png
1436 | n03-113-01.png
1437 | n06-128-00.png
1438 | n06-111-01.png
1439 | n03-106-02.png
1440 | d06-107-01.png
1441 | m06-048-00.png
1442 | n02-045-00.png
1443 | m02-087-02.png
1444 | n02-104-03.png
1445 | g07-000b-07.png
1446 | f04-061-05.png
1447 | m01-049-09.png
1448 | n01-004-04.png
1449 | g03-049-01.png
1450 | m06-098-05.png
1451 | d06-003-06.png
1452 | n01-052-00.png
1453 | p02-076-08.png
1454 | m01-110-00.png
1455 | f04-071-01.png
1456 | n04-000-00.png
1457 | n04-068-03.png
1458 | n04-092-08.png
1459 | n02-004-01.png
1460 | d04-111-03.png
1461 | f07-016-10.png
1462 | f04-071-02.png
1463 | d04-089-06.png
1464 | g04-022-06.png
1465 | m04-131-01.png
1466 | f04-083-00.png
1467 | d01-052-03.png
1468 | n01-000-05.png
1469 | d06-056-01.png
1470 | f07-019b-06.png
1471 | p02-022-07.png
1472 | d01-052-02.png
1473 | p03-012-08.png
1474 | d03-117-00.png
1475 | g01-074-08.png
1476 | n02-037-04.png
1477 | d04-053-07.png
1478 | f07-073-06.png
1479 | n03-097-05.png
1480 | d06-025-00.png
1481 | m03-114-07.png
1482 | d01-123-05.png
1483 | m01-136-00.png
1484 | n01-045-07.png
1485 | n04-092-04.png
1486 | m01-136-09.png
1487 | d07-096-01.png
1488 | n03-120-02.png
1489 | f04-057-04.png
1490 | f07-039a-01.png
1491 | m01-121-07.png
1492 | p06-069-04.png
1493 | m04-043-04.png
1494 | p03-040-03.png
1495 | p03-181-03.png
1496 | g03-049-03.png
1497 | g01-088-09.png
1498 | e06-046-00.png
1499 | f07-092b-00.png
1500 | f07-088b-00.png
1501 | g04-043-06.png
1502 | d06-111-05.png
1503 | d07-100-08.png
1504 | d06-056-00.png
1505 | p02-081-00.png
1506 | n01-020-00.png
1507 | p06-047-09.png
1508 | m04-081-02.png
1509 | g01-016-01.png
1510 | g04-048-05.png
1511 | p03-029-03.png
1512 | p01-174-02.png
1513 | n06-074-03.png
1514 | p02-144-10.png
1515 | n01-031-07.png
1516 | f07-024b-10.png
1517 | p06-069-06.png
1518 | f07-081b-05.png
1519 | m01-079-06.png
1520 | n06-111-05.png
1521 | n04-114-04.png
1522 | n03-064-01.png
1523 | g02-062-01.png
1524 | f04-039-08.png
1525 | m03-020-03.png
1526 | d01-019-03.png
1527 | e06-000-01.png
1528 | g04-072-08.png
1529 | f07-000-08.png
1530 | m01-079-03.png
1531 | n02-109-07.png
1532 | p03-163-05.png
1533 | g01-016-03.png
1534 | d04-053-04.png
1535 | n02-045-08.png
1536 | p01-155-01.png
1537 | m02-048-09.png
1538 | f07-016-01.png
1539 | g04-048-00.png
1540 | n03-097-04.png
1541 | g03-049-00.png
1542 | m04-231-00.png
1543 | e01-055-08.png
1544 | f04-079-07.png
1545 | f07-006-05.png
1546 | n02-028-02.png
1547 | d06-003-01.png
1548 | n04-048-04.png
1549 | f04-032-01.png
1550 | f04-068-03.png
1551 | n02-114-02.png
1552 | p06-042-09.png
1553 | f07-021a-04.png
1554 | e06-003-06.png
1555 | f07-024b-03.png
1556 | m01-084-06.png
1557 | d04-016-05.png
1558 | f04-083-09.png
1559 | n03-103-04.png
1560 | p06-058-03.png
1561 | n03-103-08.png
1562 | p02-090-08.png
1563 | g02-073-02.png
1564 | p02-109-01.png
1565 | f07-101b-04.png
1566 | m01-095-00.png
1567 | d06-067-07.png
1568 | g07-079a-04.png
1569 | e06-021-04.png
1570 | m01-095-03.png
1571 | f07-084a-01.png
1572 | e06-010-02.png
1573 | f04-061-02.png
1574 | c04-134-05.png
1575 | f07-016-08.png
1576 | m01-104-02.png
1577 | m04-078-06.png
1578 | n04-092-05.png
1579 | m01-121-05.png
1580 | d06-046-02.png
1581 | n01-004-03.png
1582 | m03-110-00.png
1583 | p02-022-02.png
1584 | g07-000b-06.png
1585 | f04-074-05.png
1586 | p03-181-00.png
1587 | n04-139-04.png
1588 | m03-033-07.png
1589 | p02-150-08.png
1590 | c06-011-05.png
1591 | m04-238-03.png
1592 | n03-120-07.png
1593 | d01-019-04.png
1594 | p01-155-00.png
1595 | p03-033-05.png
1596 | n06-100-08.png
1597 | p01-168-05.png
1598 | n01-036-05.png
1599 | c04-139-01.png
1600 | g01-008-01.png
1601 | e06-003-01.png
1602 | n04-114-06.png
1603 | p02-022-08.png
1604 | n01-052-04.png
1605 | g01-027-05.png
1606 | d04-012-06.png
1607 | p01-155-05.png
1608 | g02-062-06.png
1609 | d05-008-07.png
1610 | e06-053-01.png
1611 | m06-056-01.png
1612 | e06-026-08.png
1613 | n01-009-08.png
1614 | p03-012-05.png
1615 | f07-096-03.png
1616 | n06-123-07.png
1617 | d06-020-03.png
1618 | n02-049-05.png
1619 | n06-111-06.png
1620 | m03-006-02.png
1621 | f04-064-03.png
1622 | d06-025-05.png
1623 | d05-021-08.png
1624 | f07-021a-00.png
1625 | d07-093-05.png
1626 | n03-103-00.png
1627 | f07-073-04.png
1628 | c04-144-03.png
1629 | g03-040-07.png
1630 | c04-116-02.png
1631 | d06-041-06.png
1632 | m02-095-00.png
1633 | n06-140-03.png
1634 | f04-071-03.png
1635 | g01-088-05.png
1636 | m02-109-01.png
1637 | p02-027-00.png
1638 | f07-013-02.png
1639 | e06-003-04.png
1640 | d07-100-01.png
1641 | d04-047-07.png
1642 | p02-069-06.png
1643 | n02-109-00.png
1644 | f07-013-00.png
1645 | p03-189-00.png
1646 | d06-027-07.png
1647 | f04-035-01.png
1648 | f07-081a-04.png
1649 | m04-123-06.png
1650 | n04-156-06.png
1651 | m01-104-00.png
1652 | g03-043-10.png
1653 | p02-144-09.png
1654 | p06-042-08.png
1655 | g01-012-05.png
1656 | e06-033-05.png
1657 | m01-131-04.png
1658 | p03-012-01.png
1659 | d06-060-00.png
1660 | d06-027-00.png
1661 | p06-030-05.png
1662 | d03-117-03.png
1663 | p06-052-07.png
1664 | p01-168-03.png
1665 | m04-038-01.png
1666 | f07-081a-00.png
1667 | d01-016-00.png
1668 | n03-126-02.png
1669 | m03-033-02.png
1670 | n02-033-07.png
1671 | n06-082-07.png
1672 | p06-058-06.png
1673 | c06-011-02.png
1674 | g01-067-01.png
1675 | p06-104-09.png
1676 | d01-118-00.png
1677 | n02-037-02.png
1678 | d01-019-01.png
1679 | m01-125-04.png
1680 | g04-048-03.png
1681 | f07-021a-03.png
1682 | m04-030-01.png
1683 | e06-033-00.png
1684 | p06-047-08.png
1685 | d06-086-08.png
1686 | d01-080-07.png
1687 | p02-008-08.png
1688 | d04-062-04.png
1689 | n02-045-01.png
1690 | f04-032-07.png
1691 | p02-076-00.png
1692 | f07-024a-02.png
1693 | n04-084-01.png
1694 | n02-157-07.png
1695 | m06-091-03.png
1696 | p03-033-01.png
1697 | d01-123-02.png
1698 | d06-037-00.png
1699 | d05-030-07.png
1700 | d01-016-02.png
1701 | n02-062-06.png
1702 | f07-084b-07.png
1703 | f07-092b-04.png
1704 | d07-100-00.png
1705 | g03-004-00.png
1706 | m02-080-03.png
1707 | g02-065-01.png
1708 | f07-084b-08.png
1709 | p03-072-03.png
1710 | p01-147-06.png
1711 | f07-081a-10.png
1712 | d04-101-05.png
1713 | g02-065-07.png
1714 | m04-072-02.png
1715 | p06-096-04.png
1716 | g01-008-02.png
1717 | d06-060-04.png
1718 | d04-066-08.png
1719 | f04-039-01.png
1720 | n01-031-00.png
1721 | g01-025-00.png
1722 | m04-131-06.png
1723 | n02-127-03.png
1724 | p03-027-03.png
1725 | f07-024a-08.png
1726 | n02-040-07.png
1727 | f07-081b-02.png
1728 | n02-045-05.png
1729 | n04-202-07.png
1730 | d06-020-05.png
1731 | g02-065-00.png
1732 | m02-055-05.png
1733 | d04-131-04.png
1734 | n04-139-05.png
1735 | g01-074-00.png
1736 | p02-069-01.png
1737 | m04-100-03.png
1738 | d06-003-02.png
1739 | c04-134-01.png
1740 | p03-040-00.png
1741 | m02-095-02.png
1742 | n02-040-00.png
1743 | p02-101-02.png
1744 | p02-022-04.png
1745 | e06-003-00.png
1746 | p01-147-04.png
1747 | p06-047-07.png
1748 | n04-068-00.png
1749 | m06-098-04.png
1750 | g01-067-06.png
1751 | n03-038-03.png
1752 | p06-096-02.png
1753 | d05-013-02.png
1754 | d06-086-05.png
1755 | d04-016-08.png
1756 | m03-110-02.png
1757 | d06-041-04.png
1758 | f07-032a-04.png
1759 | p03-173-06.png
1760 | g04-068-00.png
1761 | d05-025-04.png
1762 | g01-070-06.png
1763 | f07-084b-00.png
1764 | n03-113-04.png
1765 | n01-052-05.png
1766 | n02-049-03.png
1767 | n03-038-05.png
1768 | d07-102-02.png
1769 | f07-002-01.png
1770 | f07-021a-02.png
1771 | f07-028a-06.png
1772 | f04-083-05.png
1773 | n04-048-06.png
1774 | d06-063-01.png
1775 | d04-012-03.png
1776 | f07-081a-01.png
1777 | m04-231-05.png
1778 | d05-013-03.png
1779 | d06-072-09.png
1780 | d01-080-04.png
1781 | f04-039-03.png
1782 | m06-091-05.png
1783 | e06-010-03.png
1784 | p06-088-01.png
1785 | m06-098-02.png
1786 | n06-140-01.png
1787 | n02-040-04.png
1788 | d07-085-06.png
1789 | g01-070-02.png
1790 | d06-072-06.png
1791 | p06-088-03.png
1792 | g01-008-00.png
1793 | c04-170-04.png
1794 | f04-032-08.png
1795 | m01-110-09.png
1796 | n02-045-07.png
1797 | g03-049-05.png
1798 | g01-045-08.png
1799 | f04-083-02.png
1800 | m06-056-00.png
1801 | m04-072-10.png
1802 | d04-058-02.png
1803 | e06-030-02.png
1804 | n06-100-00.png
1805 | n06-111-03.png
1806 | e06-000-05.png
1807 | m04-113-04.png
1808 | p06-104-08.png
1809 | d07-085-07.png
1810 | f07-073-01.png
1811 | e01-055-07.png
1812 | f04-083-03.png
1813 | e06-026-03.png
1814 | d04-058-06.png
1815 | f07-024a-05.png
1816 | p02-069-02.png
1817 | f04-074-11.png
1818 | n06-140-04.png
1819 | c06-011-03.png
1820 | g01-019-06.png
1821 | m01-110-11.png
1822 | n03-106-05.png
1823 | d01-080-06.png
1824 | g01-088-06.png
1825 | d01-118-02.png
1826 | g01-039-07.png
1827 | p03-040-01.png
1828 | d01-080-02.png
1829 | g01-027-00.png
1830 | m02-066-04.png
1831 | n02-033-03.png
1832 | m02-095-05.png
1833 | g07-079a-02.png
1834 | f07-073-00.png
1835 | m04-246-02.png
1836 | n03-103-06.png
1837 | p02-115-02.png
1838 | p01-174-04.png
1839 | f07-084b-05.png
1840 | p02-027-01.png
1841 | d04-066-02.png
1842 | f07-013-03.png
1843 | d04-032-01.png
1844 | m01-131-00.png
1845 | f07-006-06.png
1846 | d04-111-00.png
1847 | g03-000-06.png
1848 | g07-000b-03.png
1849 | f04-053-02.png
1850 | p02-081-03.png
1851 | p02-022-03.png
1852 | f07-088a-06.png
1853 | g01-027-01.png
1854 | f04-032-06.png
1855 | n04-068-07.png
1856 | m01-084-07.png
1857 | n06-082-00.png
1858 | d04-050-05.png
1859 | g01-034-07.png
1860 | n04-114-05.png
1861 | g01-016-00.png
1862 | g02-073-04.png
1863 | g01-034-02.png
1864 | d07-085-02.png
1865 | f07-032a-06.png
1866 | d06-107-02.png
1867 | m01-110-04.png
1868 | g01-031-00.png
1869 | f04-039-05.png
1870 | f07-006-02.png
1871 | g02-062-09.png
1872 | g04-048-02.png
1873 | n02-109-06.png
1874 | m06-083-04.png
1875 | n03-126-01.png
1876 | n02-037-08.png
1877 | d04-037-05.png
1878 | g03-052-08.png
1879 | d06-111-01.png
1880 | f04-064-06.png
1881 | f07-021a-06.png
1882 | m04-138-05.png
1883 | d06-100-05.png
1884 | e06-033-01.png
1885 | n04-213-04.png
1886 | n04-218-06.png
1887 | n04-075-03.png
1888 | p02-081-04.png
1889 | f04-079-04.png
1890 | d06-011-05.png
1891 | g01-019-07.png
1892 | d04-021-01.png
1893 | d05-021-05.png
1894 | n02-151-04.png
1895 | m06-091-00.png
1896 | n02-120-03.png
1897 | n01-057-08.png
1898 | n04-163-03.png
1899 | f07-069-00.png
1900 | d04-089-07.png
1901 | g01-034-06.png
1902 | f07-024b-00.png
1903 | n02-009-05.png
1904 | n06-123-04.png
1905 | f07-019b-10.png
1906 | n02-009-06.png
1907 | m04-078-07.png
1908 | g04-043-05.png
1909 | e01-055-01.png
1910 | d06-096-00.png
1911 | f07-016-09.png
1912 | f07-032a-05.png
1913 | f07-081b-04.png
1914 | e01-055-05.png
1915 | n02-157-00.png
1916 | n04-156-07.png
1917 | d05-013-04.png
1918 | n03-106-01.png
1919 | p01-174-08.png
1920 | f04-057-06.png
1921 | g03-052-10.png
1922 | d06-072-10.png
1923 | d04-053-08.png
1924 | f07-019b-08.png
1925 | f07-028a-02.png
1926 | d06-025-07.png
1927 | g01-031-08.png
1928 | f04-071-07.png
1929 | f07-096-04.png
1930 | f07-042a-07.png
1931 | f07-006-03.png
1932 | f07-081a-03.png
1933 | p03-023-00.png
1934 | m04-078-02.png
1935 | c04-150-03.png
1936 | g01-034-03.png
1937 | m06-031-05.png
1938 | n02-009-01.png
1939 | f04-079-01.png
1940 | m03-110-06.png
1941 | p03-012-09.png
1942 | g03-032-02.png
1943 | d05-040-04.png
1944 | n04-009-05.png
1945 | d04-096-06.png
1946 | f07-081a-08.png
1947 | m02-048-02.png
1948 | e06-033-09.png
1949 | p06-104-00.png
1950 | g04-043-08.png
1951 | m01-110-03.png
1952 | n04-156-03.png
1953 | d07-082-07.png
1954 | n04-213-02.png
1955 | d06-041-00.png
1956 | e06-015-00.png
1957 | m04-246-06.png
1958 | p03-027-07.png
1959 | g01-088-03.png
1960 | e06-000-03.png
1961 | n04-149-06.png
1962 | n04-139-06.png
1963 | n02-127-01.png
1964 | c04-165-00.png
1965 | d04-053-01.png
1966 | d06-100-06.png
1967 | m03-006-08.png
1968 | e06-053-02.png
1969 | d05-021-03.png
1970 | g04-022-01.png
1971 | e06-033-02.png
1972 | p02-109-03.png
1973 | g01-025-07.png
1974 | d04-047-02.png
1975 | g04-068-09.png
1976 | e06-021-08.png
1977 | m06-106-06.png
1978 | f07-076a-02.png
1979 | n04-202-08.png
1980 | m01-131-01.png
1981 | m04-123-03.png
1982 | p02-069-05.png
1983 | n04-000-02.png
1984 | g01-070-03.png
1985 | n01-000-06.png
1986 | d07-096-03.png
1987 | d04-012-01.png
1988 | d04-125-01.png
1989 | g01-027-02.png
1990 | d04-012-02.png
1991 | d06-063-06.png
1992 | d06-056-08.png
1993 | d07-093-08.png
1994 | d01-049-02.png
1995 | f07-021a-07.png
1996 | m02-080-01.png
1997 | d06-104-10.png
1998 | p06-096-08.png
1999 | n02-120-05.png
2000 | d05-030-05.png
2001 | m02-048-00.png
2002 | n04-092-00.png
2003 | p02-076-02.png
2004 | d01-049-08.png
2005 | d04-047-08.png
2006 | n01-052-09.png
2007 | n04-130-02.png
2008 | n06-128-03.png
2009 | d07-096-06.png
2010 | d05-008-04.png
2011 | m04-107-08.png
2012 | n06-100-06.png
2013 | m03-013-02.png
2014 | f07-084b-02.png
2015 | g01-027-08.png
2016 | n02-054-05.png
2017 | m06-098-01.png
2018 | f07-084a-04.png
2019 | m06-106-03.png
2020 | n03-066-01.png
2021 | n03-126-03.png
2022 | n06-074-00.png
2023 | n04-130-04.png
2024 | m02-059-04.png
2025 | m02-059-05.png
2026 | f04-079-08.png
2027 | n02-154-00.png
2028 | n06-074-04.png
2029 | f07-009-06.png
2030 | c04-139-05.png
2031 | d06-020-00.png
2032 | f07-019a-08.png
2033 | p01-174-00.png
2034 | g07-079a-05.png
2035 | c04-116-01.png
2036 | d04-066-00.png
2037 | p02-081-02.png
2038 | n03-126-04.png
2039 | m02-072-03.png
2040 | d06-076-04.png
2041 | e06-003-10.png
2042 | p02-022-01.png
2043 | d01-085-03.png
2044 | g01-039-09.png
2045 | f07-039a-00.png
2046 | g01-004-02.png
2047 | p02-069-09.png
2048 | d06-113-02.png
2049 | f07-081b-00.png
2050 | f07-019a-03.png
2051 | g01-045-01.png
2052 | g03-004-05.png
2053 | n01-052-06.png
2054 | n04-213-06.png
2055 | d05-008-05.png
2056 | g03-000-05.png
2057 | d03-117-04.png
2058 | d06-072-01.png
2059 | d07-082-06.png
2060 | f04-035-03.png
2061 | g03-064-06.png
2062 | e06-000-06.png
2063 | m01-125-03.png
2064 | p06-104-02.png
2065 | d04-053-06.png
2066 | d07-096-08.png
2067 | g02-062-00.png
2068 | m02-066-03.png
2069 | m06-076-01.png
2070 | n02-037-07.png
2071 | f07-021b-01.png
2072 | p02-139-02.png
2073 | g01-043-05.png
2074 | g07-000b-04.png
2075 | n02-033-04.png
2076 | e06-030-05.png
2077 | f07-073-03.png
2078 | m02-080-06.png
2079 | d06-063-03.png
2080 | d05-021-01.png
2081 | d06-067-08.png
2082 | d06-025-06.png
2083 | d04-101-01.png
2084 | d06-046-07.png
2085 | n03-066-07.png
2086 | n01-045-00.png
2087 | m01-049-00.png
2088 | f07-028a-01.png
2089 | n04-100-03.png
2090 | d07-102-07.png
2091 | f07-073-05.png
2092 | e06-026-04.png
2093 | n01-020-04.png
2094 | d04-032-00.png
2095 | n02-045-06.png
2096 | f07-000-07.png
2097 | p03-029-02.png
2098 | n02-009-03.png
2099 | f07-016-06.png
2100 | m01-110-06.png
2101 | d05-008-01.png
2102 | f07-000b-01.png
2103 | p02-090-06.png
2104 | n02-040-06.png
2105 | d05-008-06.png
2106 | m01-049-10.png
2107 | d06-076-01.png
2108 | p03-163-07.png
2109 | n04-149-03.png
2110 | f07-006-04.png
2111 | c04-165-07.png
2112 | p02-081-01.png
2113 | d06-025-02.png
2114 | m01-049-01.png
2115 | m02-109-00.png
2116 | n04-075-00.png
2117 | e06-015-06.png
2118 | p02-150-06.png
2119 | n02-004-04.png
2120 | f07-006-01.png
2121 | n03-120-03.png
2122 | p02-008-00.png
2123 | n01-045-08.png
2124 | d06-072-08.png
2125 | d06-041-07.png
2126 | f07-013-06.png
2127 | m06-067-02.png
2128 | n03-120-01.png
2129 | m04-072-07.png
2130 | m04-107-04.png
2131 | f04-064-02.png
2132 | n04-209-01.png
2133 | m01-131-02.png
2134 | g01-070-04.png
2135 | m04-113-03.png
2136 | d05-021-07.png
2137 | d06-027-01.png
2138 | d07-089-04.png
2139 | p02-144-11.png
2140 | p03-012-02.png
2141 | n04-209-07.png
2142 | d04-016-03.png
2143 | p06-096-05.png
2144 | f07-000-00.png
2145 | n06-123-00.png
2146 | m01-110-05.png
2147 | g01-037-06.png
2148 | n01-057-00.png
2149 | f04-053-00.png
2150 | d01-080-01.png
2151 | m03-020-02.png
2152 | e06-049-09.png
2153 | p03-189-03.png
2154 | d07-100-07.png
2155 | c04-144-00.png
2156 | f04-068-07.png
2157 | d05-013-08.png
2158 | n04-218-05.png
2159 | d06-011-03.png
2160 | g04-068-05.png
2161 | n03-106-00.png
2162 | n04-084-06.png
2163 | p01-168-04.png
2164 | p06-030-00.png
2165 | p01-147-03.png
2166 | d06-060-02.png
2167 | p02-081-08.png
2168 | m03-013-01.png
2169 | c06-011-07.png
2170 | d01-016-01.png
2171 | n02-028-07.png
2172 | f07-081a-07.png
2173 | p02-027-02.png
2174 | m04-107-09.png
2175 | c06-011-06.png
2176 | g04-068-02.png
2177 | d04-012-10.png
2178 | f07-076a-05.png
2179 | g01-067-03.png
2180 | p06-052-08.png
2181 | p06-088-05.png
2182 | f07-084b-06.png
2183 | d04-058-04.png
2184 | m02-048-04.png
2185 | g02-065-09.png
2186 | d04-016-01.png
2187 | g01-031-02.png
2188 | m06-048-03.png
2189 | f07-019a-02.png
2190 | d04-111-10.png
2191 | m04-081-06.png
2192 | n02-151-05.png
2193 | n02-104-06.png
2194 | n03-120-08.png
2195 | p03-173-00.png
2196 | p03-072-06.png
2197 | n02-157-02.png
2198 | m01-104-03.png
2199 | m06-083-07.png
2200 | n02-120-04.png
2201 | f07-069-02.png
2202 | p02-115-00.png
2203 | d07-096-00.png
2204 | f07-013-08.png
2205 | d06-086-09.png
2206 | e06-026-02.png
2207 | n06-123-01.png
2208 | f04-071-00.png
2209 | n02-062-02.png
2210 | p01-174-06.png
2211 | d07-085-03.png
2212 | m04-100-04.png
2213 | m03-013-06.png
2214 | n02-151-07.png
2215 | p06-088-09.png
2216 | n02-109-02.png
2217 | d01-080-05.png
2218 | m06-106-08.png
2219 | m01-121-03.png
2220 | m01-125-07.png
2221 | f07-028a-04.png
2222 | m03-006-07.png
2223 | p06-058-10.png
2224 | d06-063-10.png
2225 | g03-000-08.png
2226 | f07-088b-01.png
2227 | d01-098-05.png
2228 | d04-066-03.png
2229 | p03-080-00.png
2230 | d06-104-08.png
2231 | n01-009-05.png
2232 | p06-058-04.png
2233 | d06-008-03.png
2234 | f07-092a-03.png
2235 | d06-011-04.png
2236 | d06-011-07.png
2237 | p06-030-08.png
2238 | n01-009-01.png
2239 | f04-043-07.png
2240 | m01-149-04.png
2241 | d04-047-06.png
2242 | m03-114-06.png
2243 | n04-149-00.png
2244 | p02-090-04.png
2245 | e06-070-06.png
2246 | d01-104-03.png
2247 | d06-060-09.png
2248 | n02-033-02.png
2249 | f04-074-10.png
2250 | d01-118-05.png
2251 | e06-049-02.png
2252 | n04-075-04.png
2253 | p06-069-08.png
2254 | m02-095-09.png
2255 | n02-154-06.png
2256 | p02-109-05.png
2257 | n02-037-01.png
2258 | d04-028-00.png
2259 | d06-041-05.png
2260 | d05-013-00.png
2261 | g01-019-04.png
2262 | n03-106-09.png
2263 | g03-000-09.png
2264 | d07-102-04.png
2265 | m04-093-05.png
2266 | m06-106-01.png
2267 | m01-060-07.png
2268 | f04-064-04.png
2269 | g01-037-05.png
2270 | n04-084-00.png
2271 | p02-027-03.png
2272 | d06-104-07.png
2273 | d05-021-00.png
2274 | n03-113-06.png
2275 | d06-086-06.png
2276 | n04-213-00.png
2277 | c04-165-06.png
2278 | g02-065-05.png
2279 | n02-151-06.png
2280 | n04-009-02.png
2281 | m03-114-04.png
2282 | d04-101-04.png
2283 | m01-110-10.png
2284 | g07-079a-07.png
2285 | n04-163-05.png
2286 | n06-140-05.png
2287 | p06-030-07.png
2288 | f07-009-05.png
2289 | n04-130-01.png
2290 | c04-150-02.png
2291 | m03-118-03.png
2292 | n03-097-07.png
2293 | m04-113-07.png
2294 | d04-021-00.png
2295 | d03-117-07.png
2296 | m06-106-00.png
2297 | p02-109-08.png
2298 | d06-104-00.png
2299 | d04-021-05.png
2300 | d04-058-05.png
2301 | f07-088b-03.png
2302 | n04-075-06.png
2303 | m04-107-01.png
2304 | n01-000-00.png
2305 | m02-109-05.png
2306 | n02-104-04.png
2307 | d03-117-08.png
2308 | p03-040-02.png
2309 | f07-000b-03.png
2310 | f07-088a-09.png
2311 | p02-150-04.png
2312 | f04-079-02.png
2313 | f07-076a-03.png
2314 | m02-087-07.png
2315 | f07-032a-01.png
2316 | d05-008-00.png
2317 | d04-101-03.png
2318 | m02-066-00.png
2319 | m02-048-07.png
2320 | e06-070-08.png
2321 | m02-087-04.png
2322 | g01-070-09.png
2323 | g03-052-07.png
2324 | p02-008-09.png
2325 | d06-046-00.png
2326 | m01-131-03.png
2327 | m06-076-02.png
2328 | p03-189-02.png
2329 | m01-084-05.png
2330 | m06-042-00.png
2331 | d04-047-00.png
2332 | n03-097-08.png
2333 | p02-150-03.png
2334 | n03-097-03.png
2335 | p06-058-08.png
2336 | f04-074-00.png
2337 | m03-033-06.png
2338 | f07-000b-02.png
2339 | m02-112-02.png
2340 | f07-046a-04.png
2341 | n03-066-00.png
2342 | p03-181-09.png
2343 | c04-110-00.png
2344 | g01-074-06.png
2345 | m02-066-01.png
2346 | n06-119-06.png
2347 | m06-042-04.png
2348 | g03-032-06.png
2349 | n03-066-02.png
2350 | f07-000-05.png
2351 | g03-000-04.png
2352 | m06-106-04.png
2353 | n04-000-01.png
2354 | p03-029-04.png
2355 | n04-163-01.png
2356 | n06-128-06.png
2357 | m03-013-00.png
2358 | m01-079-02.png
2359 | n04-100-02.png
2360 | p02-090-00.png
2361 | d06-072-07.png
2362 | f07-016-00.png
2363 | p06-088-04.png
2364 | e06-030-09.png
2365 | m04-138-07.png
2366 | p02-081-09.png
2367 | d04-086-05.png
2368 | g01-012-02.png
2369 | c04-144-05.png
2370 | d06-096-03.png
2371 | g03-004-02.png
2372 | p03-189-01.png
2373 | m01-049-08.png
2374 | n01-004-05.png
2375 | f07-084b-04.png
2376 | p06-047-01.png
2377 | m06-031-06.png
2378 | d06-027-06.png
2379 | f04-035-04.png
2380 | m06-067-04.png
2381 | d05-030-03.png
2382 | c04-165-08.png
2383 | d04-012-05.png
2384 | d07-102-03.png
2385 | f07-002-08.png
2386 | m04-043-00.png
2387 | n02-049-01.png
2388 | f07-032a-03.png
2389 | d06-008-02.png
2390 | m01-084-04.png
2391 | c04-150-06.png
2392 | n03-038-00.png
2393 | m03-118-02.png
2394 | p01-155-03.png
2395 | f07-042a-01.png
2396 | m02-069-07.png
2397 | p02-115-01.png
2398 | p02-139-01.png
2399 | p06-096-06.png
2400 | g04-043-00.png
2401 | f07-009-00.png
2402 | n02-154-01.png
2403 | m02-080-05.png
2404 | d06-100-02.png
2405 | m02-072-02.png
2406 | p02-144-04.png
2407 | p02-008-10.png
2408 | p03-163-06.png
2409 | g03-043-05.png
2410 | f07-042a-02.png
2411 | d05-021-02.png
2412 | m01-121-01.png
2413 | n04-048-01.png
2414 | n04-107-02.png
2415 | m04-093-09.png
2416 | n04-209-03.png
2417 | g01-043-02.png
2418 | e06-030-01.png
2419 | m04-100-07.png
2420 | g01-045-03.png
2421 | n02-154-04.png
2422 | d06-063-08.png
2423 | g01-025-06.png
2424 | n03-113-05.png
2425 | p02-139-00.png
2426 | m01-095-02.png
2427 | d04-058-01.png
2428 | p06-058-00.png
2429 | n02-028-05.png
2430 | n04-163-06.png
2431 | f07-000-03.png
2432 | p03-163-03.png
2433 | d04-066-05.png
2434 | m02-072-04.png
2435 | g04-022-02.png
2436 | f04-074-03.png
2437 | n03-066-05.png
2438 | d04-012-00.png
2439 | f07-101b-02.png
2440 | m06-067-03.png
2441 | n04-060-04.png
2442 | n06-092-05.png
2443 | p02-090-05.png
2444 | m01-049-06.png
2445 | d06-107-10.png
2446 | e06-026-06.png
2447 | m06-098-03.png
2448 | d01-104-02.png
2449 | n02-114-07.png
2450 | d07-089-00.png
2451 | p02-109-07.png
2452 | f07-028a-05.png
2453 | n06-092-04.png
2454 | m02-087-05.png
2455 | d01-104-00.png
2456 | f07-092b-02.png
2457 | m04-093-07.png
2458 | n04-100-05.png
2459 | f07-092b-06.png
2460 | n02-037-05.png
2461 | n02-114-05.png
2462 | f07-076a-08.png
2463 | g01-037-07.png
2464 | n04-209-00.png
2465 | d06-008-06.png
2466 | p02-139-04.png
2467 | g01-074-01.png
2468 | d07-093-02.png
2469 | m04-078-04.png
2470 | n02-028-04.png
2471 | n06-082-05.png
2472 | p03-047-00.png
2473 | d06-003-00.png
2474 | f07-024b-06.png
2475 | f04-061-04.png
2476 | p03-163-02.png
2477 | d04-028-02.png
2478 | m02-055-07.png
2479 | f04-057-02.png
2480 | d06-113-03.png
2481 | n03-103-02.png
2482 | n04-107-00.png
2483 | e06-070-05.png
2484 | n04-100-01.png
2485 | d04-111-04.png
2486 | d04-028-06.png
2487 | n04-009-01.png
2488 | e06-000-00.png
2489 | g01-012-00.png
2490 | p02-105-03.png
2491 | f07-000-02.png
2492 | n02-054-00.png
2493 | n02-004-03.png
2494 | p01-168-07.png
2495 | m06-031-02.png
2496 | d06-008-00.png
2497 | d06-056-02.png
2498 | m01-125-01.png
2499 | n01-004-06.png
2500 | p03-012-06.png
2501 | f07-016-03.png
2502 | f07-042a-04.png
2503 | m03-020-04.png
2504 | g03-043-09.png
2505 | n01-031-03.png
2506 | d04-089-03.png
2507 | n04-068-02.png
2508 | m03-114-05.png
2509 | m04-231-01.png
2510 | d06-096-04.png
2511 | p03-181-01.png
2512 | f04-032-02.png
2513 | n04-052-08.png
2514 | f07-021b-07.png
2515 | d01-052-00.png
2516 | d05-021-04.png
2517 | m04-072-01.png
2518 | n04-000-05.png
2519 | m01-060-02.png
2520 | f07-013-04.png
2521 | p06-096-01.png
2522 | f07-069-03.png
2523 | m04-078-03.png
2524 | n02-114-00.png
2525 | f04-043-03.png
2526 | g01-074-05.png
2527 | m06-091-01.png
2528 | p06-069-10.png
2529 | p03-040-05.png
2530 | c04-144-07.png
2531 | m03-006-01.png
2532 | p01-155-02.png
2533 | m03-013-04.png
2534 | g04-048-01.png
2535 | m01-136-01.png
2536 | f07-096-00.png
2537 | g01-027-06.png
2538 | m01-060-04.png
2539 | p01-168-01.png
2540 | p02-135-05.png
2541 | e06-049-01.png
2542 | p02-000-05.png
2543 | m04-123-10.png
2544 | m04-107-00.png
2545 | d04-037-04.png
2546 | d05-013-01.png
2547 | d06-020-04.png
2548 | m02-109-04.png
2549 | n02-004-00.png
2550 | n02-009-00.png
2551 | c04-144-06.png
2552 | p06-042-04.png
2553 | p01-147-08.png
2554 | f07-092b-01.png
2555 | f07-000-04.png
2556 | e06-010-07.png
2557 | d04-047-03.png
2558 | f04-071-06.png
2559 | d04-032-05.png
2560 | n02-114-04.png
2561 | f07-073-09.png
2562 | f04-035-02.png
2563 | m02-069-03.png
2564 | m04-107-07.png
2565 | p03-033-00.png
2566 | p02-090-01.png
2567 | m04-138-06.png
2568 | m04-093-04.png
2569 | g03-043-00.png
2570 | d04-032-09.png
2571 | g03-043-07.png
2572 | n06-082-06.png
2573 | d05-040-01.png
2574 | g01-088-00.png
2575 | n06-119-01.png
2576 | d07-082-08.png
2577 | m02-048-03.png
2578 | m02-095-07.png
2579 | f07-021b-04.png
2580 | g03-032-00.png
2581 | m03-114-03.png
2582 | d06-072-03.png
2583 | d06-107-09.png
2584 | d07-102-05.png
2585 | g04-052-02.png
2586 | m01-079-01.png
2587 | f04-071-08.png
2588 | d06-086-03.png
2589 | f04-032-09.png
2590 | g01-004-04.png
2591 | p02-150-05.png
2592 | g03-000-07.png
2593 | n02-104-02.png
2594 | r03-053-00.png
2595 | g03-049-02.png
2596 | g03-043-01.png
2597 | m03-114-09.png
2598 | g01-043-06.png
2599 | g04-052-00.png
2600 | m06-056-03.png
2601 | n01-020-08.png
2602 | n06-133-02.png
2603 | p02-135-03.png
2604 | f04-083-07.png
2605 | d01-049-04.png
2606 | m02-109-08.png
2607 | m02-087-01.png
2608 | p01-155-09.png
2609 | f07-024b-02.png
2610 | m02-048-01.png
2611 | d06-067-06.png
2612 | m03-114-00.png
2613 | n02-127-00.png
2614 | d06-037-06.png
2615 | n04-202-05.png
2616 | n06-092-08.png
2617 | d04-096-03.png
2618 | n04-084-02.png
2619 | e06-030-08.png
2620 | p03-040-06.png
2621 | d01-049-00.png
2622 | m03-118-01.png
2623 | m02-069-02.png
2624 | d04-028-03.png
2625 | d06-003-04.png
2626 | d07-089-08.png
2627 | f07-092a-00.png
2628 | m04-138-02.png
2629 | f07-019b-00.png
2630 | m04-093-06.png
2631 | d01-118-01.png
2632 | d04-131-06.png
2633 | d04-111-01.png
2634 | f07-028a-03.png
2635 | d04-096-04.png
2636 | m03-020-05.png
2637 | f07-088a-00.png
2638 | f07-092a-01.png
2639 | m06-083-05.png
2640 | f07-046a-00.png
2641 | g01-037-09.png
2642 | n01-052-03.png
2643 | f07-084a-07.png
2644 | d04-086-01.png
2645 | d06-067-04.png
2646 | d06-096-08.png
2647 | m01-079-05.png
2648 | f07-084a-02.png
2649 | p03-029-05.png
2650 | d01-052-04.png
2651 | n02-062-03.png
2652 | d01-085-06.png
2653 | e06-070-04.png
2654 | p02-022-00.png
2655 | m04-246-05.png
2656 | e06-030-10.png
2657 | n01-009-04.png
2658 | n01-031-02.png
2659 | n03-097-02.png
2660 | e06-003-05.png
2661 | f04-057-08.png
2662 | c04-170-01.png
2663 | m01-049-03.png
2664 | g01-067-07.png
2665 | e06-021-06.png
2666 | f07-032a-09.png
2667 | m04-246-03.png
2668 | n04-218-07.png
2669 | d07-089-05.png
2670 | f07-019a-06.png
2671 | d01-098-02.png
2672 | m06-106-09.png
2673 | n04-048-00.png
2674 | d04-096-00.png
2675 | f07-076a-07.png
2676 | e06-049-00.png
2677 | d04-062-01.png
2678 | f07-013-10.png
2679 | d06-104-05.png
2680 | d05-030-00.png
2681 | e06-030-03.png
2682 | f04-057-01.png
2683 | p03-027-06.png
2684 | p06-047-05.png
2685 | n04-218-02.png
2686 | p02-139-03.png
2687 | p03-047-09.png
2688 | n06-133-03.png
2689 | g03-000-03.png
2690 | f07-092a-08.png
2691 | m03-033-03.png
2692 | n06-082-03.png
2693 | m01-149-02.png
2694 | d06-067-02.png
2695 | d04-125-04.png
2696 | p03-027-04.png
2697 | e06-053-04.png
2698 | e06-070-07.png
2699 | d04-062-00.png
2700 | d04-053-10.png
2701 | m01-136-05.png
2702 | g01-019-03.png
2703 | p06-104-07.png
2704 | d07-089-09.png
2705 | n02-151-08.png
2706 | f04-083-04.png
2707 | d04-096-05.png
2708 | d06-063-09.png
2709 | m04-107-03.png
2710 | p02-022-06.png
2711 | g01-067-02.png
2712 | d06-020-08.png
2713 | d04-096-02.png
2714 | f04-032-04.png
2715 | c04-134-04.png
2716 | m01-095-01.png
2717 | f04-053-03.png
2718 | p02-115-06.png
2719 | g01-019-09.png
2720 | m04-238-01.png
2721 | p02-115-07.png
2722 | d04-032-10.png
2723 | n03-064-05.png
2724 | d07-100-09.png
2725 | f07-021b-03.png
2726 | f07-009-02.png
2727 | g02-065-08.png
2728 | p03-173-04.png
2729 | d06-060-01.png
2730 | d05-013-06.png
2731 | p01-155-07.png
2732 | p03-163-04.png
2733 | p02-144-01.png
2734 | d06-086-04.png
2735 | m01-136-10.png
2736 | n06-074-02.png
2737 | p02-069-03.png
2738 | m02-112-00.png
2739 | g03-064-04.png
2740 | f07-028a-00.png
2741 | m02-112-04.png
2742 | n03-103-01.png
2743 | p06-052-03.png
2744 | f04-079-00.png
2745 | m02-087-03.png
2746 | n04-202-04.png
2747 | d04-037-06.png
2748 | m01-049-02.png
2749 | d06-076-05.png
2750 | m04-113-08.png
2751 | e06-026-07.png
2752 | g03-000-00.png
2753 | m01-095-06.png
2754 | n06-133-08.png
2755 | g01-034-05.png
2756 | m06-048-04.png
2757 | n01-031-01.png
2758 | d07-085-05.png
2759 | n02-104-00.png
2760 | g01-031-07.png
2761 | g03-040-06.png
2762 | m04-113-06.png
2763 | m03-033-01.png
2764 | n02-062-04.png
2765 | d01-104-08.png
2766 | p03-080-04.png
2767 | p06-096-09.png
2768 | f07-019b-04.png
2769 | f07-024b-08.png
2770 | m06-091-02.png
2771 | r03-053-03.png
2772 | d04-016-07.png
2773 | n04-163-08.png
2774 | m01-125-05.png
2775 | n01-036-03.png
2776 | m04-081-07.png
2777 | d07-089-01.png
2778 | f07-088a-02.png
2779 | f07-002-02.png
2780 | g02-073-01.png
2781 | m01-084-00.png
2782 | p06-042-06.png
2783 | f04-087-01.png
2784 | g07-000b-08.png
2785 | g03-004-04.png
2786 | f07-002-07.png
2787 | f04-071-09.png
2788 | d07-082-04.png
2789 | n02-033-00.png
2790 | g03-052-02.png
2791 | d04-066-01.png
2792 | n03-038-02.png
2793 | n06-133-00.png
2794 | e06-033-06.png
2795 | n04-048-05.png
2796 | g01-039-01.png
2797 | g01-043-03.png
2798 | n04-209-06.png
2799 | p01-168-02.png
2800 | m04-131-05.png
2801 | n01-000-01.png
2802 | p02-008-07.png
2803 | e06-046-02.png
2804 | f07-046a-09.png
2805 | g03-043-03.png
2806 | p06-052-05.png
2807 | f04-043-01.png
2808 | n04-048-08.png
2809 | f07-042a-03.png
2810 | p02-135-08.png
2811 | n03-106-04.png
2812 | p06-042-02.png
2813 | d06-076-00.png
2814 | f07-084b-01.png
2815 | f07-039a-04.png
2816 | m04-081-03.png
2817 | n01-057-09.png
2818 | g01-039-04.png
2819 | p01-147-07.png
2820 | p06-042-01.png
2821 | m02-055-02.png
2822 | c04-165-03.png
2823 | n06-123-10.png
2824 | n04-139-02.png
2825 | m02-055-06.png
2826 | n02-028-03.png
2827 | p06-042-10.png
2828 | d04-028-01.png
2829 | m03-114-02.png
2830 | p06-104-06.png
2831 | f07-046a-06.png
2832 | n02-033-06.png
2833 | e06-015-05.png
2834 | f07-009-03.png
2835 | d06-003-05.png
2836 | f07-092a-04.png
2837 | g01-067-05.png
2838 | f07-092a-02.png
2839 | g07-079a-06.png
2840 | p02-150-07.png
2841 | f07-076a-10.png
2842 | m06-091-04.png
2843 | c04-116-00.png
2844 | f07-021b-09.png
2845 | m04-238-02.png
2846 | m02-055-03.png
2847 | d06-011-09.png
2848 | f07-088a-08.png
2849 | g01-019-05.png
2850 | m06-048-02.png
2851 | f07-002-04.png
2852 | n06-123-03.png
2853 | f07-076a-00.png
2854 | g04-052-04.png
2855 | n04-048-02.png
2856 | f07-019a-04.png
2857 | n02-120-00.png
2858 | f07-088a-05.png
2859 | p02-115-08.png
2860 | e06-030-00.png
2861 | m03-110-05.png
2862 | d04-021-03.png
2863 | n02-120-07.png
2864 | p02-000-03.png
2865 | p03-181-07.png
2866 | n06-128-02.png
2867 | n06-119-00.png
2868 | p06-052-06.png
2869 | f07-000b-00.png
2870 | p02-076-03.png
2871 | d06-111-06.png
2872 | f04-068-05.png
2873 | n06-082-02.png
2874 | m04-231-06.png
2875 | m01-060-05.png
2876 | m01-060-03.png
2877 | p06-030-01.png
2878 | g03-052-09.png
2879 | d07-085-04.png
2880 | n04-130-03.png
2881 | n02-028-08.png
2882 | f07-019b-03.png
2883 | d01-123-04.png
2884 | n03-120-06.png
2885 | g04-068-06.png
2886 | m03-006-03.png
2887 | g04-072-06.png
2888 | g07-079a-00.png
2889 | e06-046-04.png
2890 | g01-016-02.png
2891 | d06-056-05.png
2892 | n04-092-06.png
2893 | n04-075-05.png
2894 | f04-079-05.png
2895 | m02-072-06.png
2896 | f07-019b-02.png
2897 | n01-036-04.png
2898 | m01-125-00.png
2899 | n04-060-02.png
2900 | n06-128-04.png
2901 | c04-134-02.png
2902 | n04-114-03.png
2903 | f04-074-09.png
2904 | m06-048-07.png
2905 | p03-181-08.png
2906 | p06-096-10.png
2907 | d06-037-01.png
2908 | e06-000-04.png
2909 | f04-074-08.png
2910 | n04-218-03.png
2911 | d04-125-02.png
2912 | n03-106-11.png
2913 | d06-113-04.png
2914 | m04-043-02.png
2915 | d01-104-01.png
2916 | 


--------------------------------------------------------------------------------
/iam/train.gc:
--------------------------------------------------------------------------------
  1 | a01-000u.png
  2 | a01-000x.png
  3 | a01-003.png
  4 | a01-003u.png
  5 | a01-003x.png
  6 | a01-007.png
  7 | a01-007u.png
  8 | a01-007x.png
  9 | a01-011.png
 10 | a01-011u.png
 11 | a01-011x.png
 12 | a01-014.png
 13 | a01-014u.png
 14 | a01-014x.png
 15 | a01-020.png
 16 | a01-020u.png
 17 | a01-020x.png
 18 | a01-026.png
 19 | a01-026u.png
 20 | a01-026x.png
 21 | a01-030.png
 22 | a01-030u.png
 23 | a01-030x.png
 24 | a01-038.png
 25 | a01-038x.png
 26 | a01-043.png
 27 | a01-043u.png
 28 | a01-043x.png
 29 | a01-049.png
 30 | a01-049u.png
 31 | a01-049x.png
 32 | a01-053.png
 33 | a01-053u.png
 34 | a01-053x.png
 35 | a01-058.png
 36 | a01-058u.png
 37 | a01-058x.png
 38 | a01-063.png
 39 | a01-063u.png
 40 | a01-063x.png
 41 | a01-068u.png
 42 | a01-072u.png
 43 | a01-072x.png
 44 | a01-077.png
 45 | a01-077u.png
 46 | a01-082u.png
 47 | a01-087.png
 48 | a01-087u.png
 49 | a01-091.png
 50 | a01-091u.png
 51 | a01-096u.png
 52 | a01-102.png
 53 | a01-102u.png
 54 | a01-107.png
 55 | a01-107u.png
 56 | a01-113.png
 57 | a01-113u.png
 58 | a01-117.png
 59 | a01-117u.png
 60 | a01-122.png
 61 | a01-122u.png
 62 | a01-128.png
 63 | a01-128u.png
 64 | a01-132.png
 65 | a01-132u.png
 66 | a01-132x.png
 67 | a02-000.png
 68 | a02-004.png
 69 | a02-008.png
 70 | a02-012.png
 71 | a02-017.png
 72 | a02-020.png
 73 | a02-024.png
 74 | a02-027.png
 75 | a02-032.png
 76 | a02-037.png
 77 | a02-042.png
 78 | a02-046.png
 79 | a02-050.png
 80 | a02-053.png
 81 | a02-057.png
 82 | a02-062.png
 83 | a02-067.png
 84 | a02-072.png
 85 | a02-078.png
 86 | a02-082.png
 87 | a02-086.png
 88 | a02-090.png
 89 | a02-093.png
 90 | a02-098.png
 91 | a02-102.png
 92 | a02-106.png
 93 | a02-111.png
 94 | a02-116.png
 95 | a02-120.png
 96 | a02-124.png
 97 | a03-006.png
 98 | a03-009.png
 99 | a03-011.png
100 | a03-014.png
101 | a03-017.png
102 | a03-020.png
103 | a03-023.png
104 | a03-027.png
105 | a03-030.png
106 | a03-034.png
107 | a03-037.png
108 | a03-040.png
109 | a03-043.png
110 | a03-047.png
111 | a03-050.png
112 | a03-054.png
113 | a03-059.png
114 | a03-063.png
115 | a03-066.png
116 | a03-071.png
117 | a03-073.png
118 | a03-080.png
119 | a03-089.png
120 | a04-000.png
121 | a04-003.png
122 | a04-006.png
123 | a04-010.png
124 | a04-015.png
125 | a04-019.png
126 | a04-023.png
127 | a04-027.png
128 | a04-031.png
129 | a04-035.png
130 | a04-039.png
131 | a04-043.png
132 | a04-047.png
133 | a04-050.png
134 | a04-054.png
135 | a04-059.png
136 | a04-066.png
137 | a04-069.png
138 | a04-072.png
139 | a04-081.png
140 | a04-085.png
141 | a04-089.png
142 | a04-092.png
143 | a04-096.png
144 | a04-099.png
145 | a04-103.png
146 | a05-000.png
147 | a05-004.png
148 | a05-009.png
149 | a05-013.png
150 | a05-017.png
151 | a05-022.png
152 | a05-025.png
153 | a05-029.png
154 | a05-039.png
155 | a05-044.png
156 | a05-048.png
157 | a05-053.png
158 | a05-058.png
159 | a05-062.png
160 | a05-069.png
161 | a05-073.png
162 | a05-080.png
163 | a05-084.png
164 | a05-089.png
165 | a05-094.png
166 | a05-099.png
167 | a05-104.png
168 | a05-108.png
169 | a05-113.png
170 | a05-116.png
171 | a05-121.png
172 | a05-125.png
173 | a06-000.png
174 | a06-004.png
175 | a06-008.png
176 | a06-014.png
177 | a06-020.png
178 | a06-025.png
179 | a06-036.png
180 | a06-039.png
181 | a06-044.png
182 | a06-051.png
183 | a06-057.png
184 | a06-064.png
185 | a06-070.png
186 | a06-075.png
187 | a06-090.png
188 | a06-095.png
189 | a06-100.png
190 | a06-110.png
191 | a06-114.png
192 | a06-119.png
193 | a06-124.png
194 | a06-128.png
195 | a06-134.png
196 | a06-141.png
197 | a06-147.png
198 | a06-152.png
199 | a06-157.png
200 | b01-000.png
201 | b01-004.png
202 | b01-009.png
203 | b01-014.png
204 | b01-018.png
205 | b01-023.png
206 | b01-027.png
207 | b01-033.png
208 | b01-038.png
209 | b01-044.png
210 | b01-049.png
211 | b01-053.png
212 | b01-057.png
213 | b01-062.png
214 | b01-073.png
215 | b01-079.png
216 | b01-084.png
217 | b01-089.png
218 | b01-094.png
219 | b01-113.png
220 | b01-118.png
221 | b01-122.png
222 | b01-127.png
223 | b01-132.png
224 | b01-136.png
225 | b02-013.png
226 | b02-035.png
227 | b02-045.png
228 | b02-097.png
229 | b02-102.png
230 | b02-105.png
231 | b03-025.png
232 | b03-087.png
233 | b03-092.png
234 | b03-098.png
235 | b03-104.png
236 | b03-109.png
237 | b03-114.png
238 | b04-000.png
239 | b04-004.png
240 | b04-010.png
241 | b04-020.png
242 | b04-026.png
243 | b04-034.png
244 | b04-040.png
245 | b04-047.png
246 | b04-054.png
247 | b04-060.png
248 | b04-066.png
249 | b04-075.png
250 | b04-081.png
251 | b04-089.png
252 | b04-096.png
253 | b04-103.png
254 | b04-107.png
255 | b04-116.png
256 | b04-121.png
257 | b04-128.png
258 | b04-134.png
259 | b04-140.png
260 | b04-147.png
261 | b04-154.png
262 | b04-162.png
263 | b04-169.png
264 | b04-175.png
265 | b04-181.png
266 | b04-187.png
267 | b04-195.png
268 | b04-202.png
269 | b04-208.png
270 | b05-017.png
271 | b05-032.png
272 | b05-038.png
273 | b05-042.png
274 | b05-045.png
275 | b05-050.png
276 | b05-055.png
277 | b05-058.png
278 | b05-062.png
279 | b05-067.png
280 | b05-071.png
281 | b05-074.png
282 | b05-079.png
283 | b05-083.png
284 | b05-088.png
285 | b05-098.png
286 | b06-000.png
287 | b06-004.png
288 | b06-008.png
289 | b06-012.png
290 | b06-015.png
291 | b06-019.png
292 | b06-023.png
293 | b06-027.png
294 | b06-032.png
295 | b06-036.png
296 | b06-042.png
297 | b06-045.png
298 | b06-049.png
299 | b06-053.png
300 | b06-056.png
301 | b06-059.png
302 | b06-064.png
303 | b06-068.png
304 | b06-071.png
305 | b06-075.png
306 | b06-079.png
307 | b06-082.png
308 | b06-087.png
309 | b06-090.png
310 | b06-093.png
311 | b06-097.png
312 | b06-100.png
313 | b06-110.png
314 | c01-009.png
315 | c01-014.png
316 | c01-066.png
317 | c02-000.png
318 | c02-003.png
319 | c02-007.png
320 | c02-012.png
321 | c02-017.png
322 | c02-022.png
323 | c02-026.png
324 | c02-030.png
325 | c02-035.png
326 | c02-049.png
327 | c02-056.png
328 | c02-059.png
329 | c02-078.png
330 | c02-082.png
331 | c02-089.png
332 | c03-000a.png
333 | c03-000b.png
334 | c03-000c.png
335 | c03-000d.png
336 | c03-000e.png
337 | c03-000f.png
338 | c03-003a.png
339 | c03-003b.png
340 | c03-003c.png
341 | c03-003d.png
342 | c03-003e.png
343 | c03-003f.png
344 | c03-007.png
345 | c03-007a.png
346 | c03-007b.png
347 | c03-007c.png
348 | c03-007d.png
349 | c03-007e.png
350 | c03-007f.png
351 | c03-016a.png
352 | c03-016b.png
353 | c03-016c.png
354 | c03-016d.png
355 | c03-016e.png
356 | c03-021a.png
357 | c03-021b.png
358 | c03-021c.png
359 | c03-021d.png
360 | c03-021e.png
361 | c03-021f.png
362 | c03-081a.png
363 | c03-081b.png
364 | c03-081c.png
365 | c03-081d.png
366 | c03-081e.png
367 | c03-081f.png
368 | c03-084a.png
369 | c03-084b.png
370 | c03-084c.png
371 | c03-084d.png
372 | c03-084e.png
373 | c03-084f.png
374 | c03-087a.png
375 | c03-087b.png
376 | c03-087c.png
377 | c03-087d.png
378 | c03-087e.png
379 | c03-087f.png
380 | c03-094a.png
381 | c03-094b.png
382 | c03-094c.png
383 | c03-094d.png
384 | c03-094e.png
385 | c03-094f.png
386 | c03-096a.png
387 | c03-096b.png
388 | c03-096c.png
389 | c03-096d.png
390 | c03-096e.png
391 | c03-096f.png
392 | c04-000.png
393 | c04-004.png
394 | c04-008.png
395 | c04-013.png
396 | c04-017.png
397 | c04-023.png
398 | c04-028.png
399 | c04-035.png
400 | c04-039.png
401 | c04-044.png
402 | c04-050.png
403 | c04-056.png
404 | c04-061.png
405 | c04-066.png
406 | c04-075.png
407 | c04-080.png
408 | c04-089.png
409 | c04-098.png
410 | c04-105.png
411 | c04-122.png
412 | c04-128.png
413 | c06-000.png
414 | c06-005.png
415 | c06-014.png
416 | c06-020.png
417 | c06-027.png
418 | c06-031.png
419 | c06-039.png
420 | c06-043.png
421 | c06-047.png
422 | c06-052.png
423 | c06-058.png
424 | c06-076.png
425 | c06-080.png
426 | c06-087.png
427 | c06-091.png
428 | c06-095.png
429 | c06-100.png
430 | c06-103.png
431 | c06-106.png
432 | c06-111.png
433 | c06-116.png
434 | c06-128.png
435 | c06-138.png
436 | e01-022.png
437 | e01-025.png
438 | e01-029.png
439 | e01-032.png
440 | e01-035.png
441 | e01-050.png
442 | e01-059.png
443 | e01-062.png
444 | e01-081.png
445 | e01-086.png
446 | e01-092.png
447 | e01-102.png
448 | e01-107.png
449 | e01-113.png
450 | e01-119.png
451 | e02-000.png
452 | e02-014.png
453 | e02-018.png
454 | e02-021.png
455 | e02-025.png
456 | e02-074.png
457 | e02-082.png
458 | e02-086.png
459 | e02-091.png
460 | e02-094.png
461 | e02-100.png
462 | e02-117.png
463 | e04-000.png
464 | e04-004.png
465 | e04-008.png
466 | e04-011.png
467 | e04-015.png
468 | e04-019.png
469 | e04-022.png
470 | e04-026.png
471 | e04-030.png
472 | e04-034.png
473 | e04-038.png
474 | e04-043.png
475 | e04-052.png
476 | e04-058.png
477 | e04-062.png
478 | e04-068.png
479 | e04-079.png
480 | e04-083.png
481 | e04-086.png
482 | e04-091.png
483 | e04-095.png
484 | e04-099.png
485 | e04-103.png
486 | e04-109.png
487 | e04-114.png
488 | e04-119.png
489 | e04-124.png
490 | e04-127.png
491 | e04-132.png
492 | e07-000.png
493 | e07-003.png
494 | e07-007.png
495 | e07-012.png
496 | e07-061.png
497 | e07-066.png
498 | e07-072.png
499 | e07-076.png
500 | e07-079.png
501 | e07-083.png
502 | e07-086.png
503 | e07-090.png
504 | e07-094.png
505 | e07-098.png
506 | e07-101.png
507 | e07-105.png
508 | e07-108.png
509 | e07-112.png
510 | f01-053.png
511 | f01-058.png
512 | f01-066.png
513 | f01-070.png
514 | f01-075.png
515 | f01-081.png
516 | f01-085.png
517 | f01-135.png
518 | f01-139.png
519 | f01-143.png
520 | f01-147.png
521 | f02-000.png
522 | f02-003.png
523 | f02-017.png
524 | f02-020.png
525 | f02-030.png
526 | f02-033.png
527 | f02-036.png
528 | f02-038.png
529 | f02-040.png
530 | f02-044.png
531 | f02-076.png
532 | f03-169.png
533 | f03-174.png
534 | f03-182.png
535 | f03-191.png
536 | f04-004.png
537 | f04-007.png
538 | f04-011.png
539 | f04-016.png
540 | f04-020.png
541 | f04-024.png
542 | f04-028.png
543 | g01-000.png
544 | g04-003.png
545 | g04-007.png
546 | g04-011.png
547 | g04-014.png
548 | g04-017.png
549 | g04-026.png
550 | g04-032.png
551 | g04-077.png
552 | g04-081.png
553 | g04-088.png
554 | g04-092.png
555 | g04-095.png
556 | g04-098.png
557 | g04-101.png
558 | g04-104.png
559 | g04-108.png
560 | g05-087.png
561 | g05-094.png
562 | g05-098.png
563 | g06-011a.png
564 | g06-011b.png
565 | g06-011c.png
566 | g06-011e.png
567 | g06-011f.png
568 | g06-011g.png
569 | g06-011h.png
570 | g06-011i.png
571 | g06-011j.png
572 | g06-011k.png
573 | g06-011l.png
574 | g06-011m.png
575 | g06-018a.png
576 | g06-018b.png
577 | g06-018c.png
578 | g06-018d.png
579 | g06-018e.png
580 | g06-018f.png
581 | g06-018g.png
582 | g06-018h.png
583 | g06-018i.png
584 | g06-018j.png
585 | g06-018k.png
586 | g06-018l.png
587 | g06-018m.png
588 | g06-026a.png
589 | g06-026b.png
590 | g06-026c.png
591 | g06-026d.png
592 | g06-026e.png
593 | g06-026f.png
594 | g06-026g.png
595 | g06-026h.png
596 | g06-026i.png
597 | g06-026j.png
598 | g06-026k.png
599 | g06-026l.png
600 | g06-026m.png
601 | g06-031a.png
602 | g06-031b.png
603 | g06-031c.png
604 | g06-031d.png
605 | g06-031e.png
606 | g06-031f.png
607 | g06-031g.png
608 | g06-031h.png
609 | g06-031i.png
610 | g06-031j.png
611 | g06-031k.png
612 | g06-031l.png
613 | g06-031m.png
614 | g06-037b.png
615 | g06-037c.png
616 | g06-037d.png
617 | g06-037e.png
618 | g06-037f.png
619 | g06-037g.png
620 | g06-037h.png
621 | g06-037i.png
622 | g06-037j.png
623 | g06-037k.png
624 | g06-037l.png
625 | g06-037m.png
626 | g06-042a.png
627 | g06-042b.png
628 | g06-042c.png
629 | g06-042d.png
630 | g06-042e.png
631 | g06-042f.png
632 | g06-042g.png
633 | g06-042h.png
634 | g06-042i.png
635 | g06-042j.png
636 | g06-042k.png
637 | g06-042l.png
638 | g06-042m.png
639 | g06-045a.png
640 | g06-045b.png
641 | g06-045c.png
642 | g06-045d.png
643 | g06-045e.png
644 | g06-045f.png
645 | g06-045g.png
646 | g06-045h.png
647 | g06-045i.png
648 | g06-045j.png
649 | g06-045k.png
650 | g06-045l.png
651 | g06-045m.png
652 | g06-047a.png
653 | g06-047b.png
654 | g06-047c.png
655 | g06-047d.png
656 | g06-047e.png
657 | g06-047f.png
658 | g06-047g.png
659 | g06-047h.png
660 | g06-047i.png
661 | g06-047j.png
662 | g06-047k.png
663 | g06-047l.png
664 | g06-047m.png
665 | g06-050a.png
666 | g06-050b.png
667 | g06-050c.png
668 | g06-050d.png
669 | g06-050e.png
670 | g06-050f.png
671 | g06-050g.png
672 | g06-050h.png
673 | g06-050i.png
674 | g06-050j.png
675 | g06-050k.png
676 | g06-050l.png
677 | g06-050m.png
678 | h04-000.png
679 | h04-007.png
680 | h04-021.png
681 | h04-025.png
682 | h05-012.png
683 | k04-017.png
684 | k04-022.png
685 | n02-146.png
686 | p06-242.png
687 | p06-248.png
688 | r02-000.png
689 | r02-003.png
690 | r02-006.png
691 | r02-010.png
692 | r02-013.png
693 | r02-038.png
694 | r02-054.png
695 | r02-060.png
696 | r02-065.png
697 | r02-070.png
698 | r02-078.png
699 | r02-081.png
700 | r02-089.png
701 | r02-109.png
702 | r02-117.png
703 | r02-127.png
704 | r02-131.png
705 | r02-137.png
706 | r02-146.png
707 | r03-002.png
708 | r03-007.png
709 | r03-026.png
710 | r03-030.png
711 | r03-035.png
712 | r03-050.png
713 | r03-056.png
714 | r03-065.png
715 | r03-084.png
716 | r03-090.png
717 | r03-096.png
718 | r03-110.png
719 | r03-115.png
720 | r06-000.png
721 | r06-003.png
722 | r06-007.png
723 | r06-011.png
724 | r06-018.png
725 | r06-022.png
726 | r06-027.png
727 | r06-035.png
728 | r06-041.png
729 | r06-044.png
730 | r06-049.png
731 | r06-053.png
732 | r06-057.png
733 | r06-062.png
734 | r06-066.png
735 | r06-070.png
736 | r06-076.png
737 | r06-090.png
738 | r06-097.png
739 | r06-103.png
740 | r06-106.png
741 | r06-111.png
742 | r06-115.png
743 | r06-121.png
744 | r06-126.png
745 | r06-130.png
746 | r06-137.png
747 | r06-143.png
748 | 


--------------------------------------------------------------------------------
/iam/val.gc:
--------------------------------------------------------------------------------
  1 | c04-156.png
  2 | c04-160.png
  3 | c06-083.png
  4 | d01-024.png
  5 | d01-056.png
  6 | d01-060.png
  7 | d03-112.png
  8 | d04-005.png
  9 | d04-008.png
 10 | d04-071.png
 11 | d04-075.png
 12 | d04-081.png
 13 | d04-117.png
 14 | d04-121.png
 15 | d06-000.png
 16 | d06-015.png
 17 | d06-030.png
 18 | d06-050.png
 19 | d06-082.png
 20 | d06-091.png
 21 | e01-014.png
 22 | e01-018.png
 23 | e06-006.png
 24 | e06-037.png
 25 | e06-041.png
 26 | f04-046.png
 27 | f04-049.png
 28 | f04-093.png
 29 | f04-096.png
 30 | f04-100.png
 31 | f07-028b.png
 32 | f07-032b.png
 33 | f07-036.png
 34 | f07-039b.png
 35 | f07-042b.png
 36 | f07-046b.png
 37 | g01-083.png
 38 | g03-016.png
 39 | g03-058.png
 40 | g04-036.png
 41 | g04-039.png
 42 | g04-055.png
 43 | g04-060.png
 44 | g04-063.png
 45 | h01-004.png
 46 | m01-090.png
 47 | m01-115.png
 48 | m01-160.png
 49 | m02-052.png
 50 | m02-075.png
 51 | m02-083.png
 52 | m02-090.png
 53 | m02-102.png
 54 | m02-106.png
 55 | m03-062.png
 56 | m03-095.png
 57 | m04-000.png
 58 | m04-007.png
 59 | m04-012.png
 60 | m04-019.png
 61 | m04-024.png
 62 | m04-145.png
 63 | m04-152.png
 64 | m04-164.png
 65 | m04-180.png
 66 | m04-190.png
 67 | m04-200.png
 68 | m04-209.png
 69 | m04-216.png
 70 | m04-222.png
 71 | m04-251.png
 72 | m06-019.png
 73 | n02-000.png
 74 | n02-016.png
 75 | n02-082a.png
 76 | n02-098.png
 77 | n03-079.png
 78 | n03-082.png
 79 | n03-091.png
 80 | n04-015.png
 81 | n04-022.png
 82 | n04-031.png
 83 | n04-039.png
 84 | n04-044.png
 85 | n04-171.png
 86 | n04-183.png
 87 | n04-190.png
 88 | n04-195.png
 89 | n06-148.png
 90 | n06-156.png
 91 | n06-163.png
 92 | n06-169.png
 93 | n06-175.png
 94 | n06-182.png
 95 | n06-186.png
 96 | n06-194.png
 97 | n06-201.png
 98 | p02-017.png
 99 | p02-121.png
100 | p02-127.png
101 | p02-131.png
102 | p02-155.png
103 | p03-004.png
104 | p03-009.png
105 | p03-057.png
106 | p03-069.png
107 | p03-087.png
108 | p03-096.png
109 | p03-103.png
110 | p03-112.png
111 | p03-121.png
112 | p03-135.png
113 | p03-142.png
114 | p03-151.png
115 | p03-158.png
116 | p03-185.png
117 | 


--------------------------------------------------------------------------------
/iam/val.ln:
--------------------------------------------------------------------------------
  1 | c04-156-06.png
  2 | c04-156-04.png
  3 | c04-156-00.png
  4 | c04-156-03.png
  5 | c04-156-01.png
  6 | c04-156-05.png
  7 | c04-156-02.png
  8 | c04-160-07.png
  9 | c04-160-06.png
 10 | c04-160-05.png
 11 | c04-160-03.png
 12 | c04-160-08.png
 13 | c04-160-09.png
 14 | c04-160-00.png
 15 | c04-160-02.png
 16 | c04-160-01.png
 17 | c04-160-04.png
 18 | c06-083-06.png
 19 | c06-083-02.png
 20 | c06-083-00.png
 21 | c06-083-01.png
 22 | c06-083-03.png
 23 | c06-083-07.png
 24 | c06-083-05.png
 25 | c06-083-04.png
 26 | c06-083-08.png
 27 | d01-024-05.png
 28 | d01-024-04.png
 29 | d01-024-00.png
 30 | d01-024-07.png
 31 | d01-024-02.png
 32 | d01-024-08.png
 33 | d01-024-01.png
 34 | d01-024-06.png
 35 | d01-024-03.png
 36 | d01-056-00.png
 37 | d01-056-03.png
 38 | d01-056-02.png
 39 | d01-056-07.png
 40 | d01-056-09.png
 41 | d01-056-06.png
 42 | d01-056-08.png
 43 | d01-056-04.png
 44 | d01-056-05.png
 45 | d01-056-01.png
 46 | d01-060-09.png
 47 | d01-060-08.png
 48 | d01-060-00.png
 49 | d01-060-03.png
 50 | d01-060-01.png
 51 | d01-060-06.png
 52 | d01-060-07.png
 53 | d01-060-05.png
 54 | d01-060-02.png
 55 | d01-060-04.png
 56 | d03-112-06.png
 57 | d03-112-03.png
 58 | d03-112-04.png
 59 | d03-112-07.png
 60 | d03-112-01.png
 61 | d03-112-02.png
 62 | d03-112-05.png
 63 | d03-112-00.png
 64 | d03-112-08.png
 65 | d04-005-01.png
 66 | d04-005-00.png
 67 | d04-005-02.png
 68 | d04-005-05.png
 69 | d04-005-06.png
 70 | d04-005-07.png
 71 | d04-005-03.png
 72 | d04-005-04.png
 73 | d04-008-05.png
 74 | d04-008-03.png
 75 | d04-008-02.png
 76 | d04-008-04.png
 77 | d04-008-01.png
 78 | d04-008-00.png
 79 | d04-071-06.png
 80 | d04-071-04.png
 81 | d04-071-03.png
 82 | d04-071-00.png
 83 | d04-071-01.png
 84 | d04-071-02.png
 85 | d04-071-07.png
 86 | d04-071-05.png
 87 | d04-075-04.png
 88 | d04-075-09.png
 89 | d04-075-03.png
 90 | d04-075-07.png
 91 | d04-075-02.png
 92 | d04-075-00.png
 93 | d04-075-06.png
 94 | d04-075-05.png
 95 | d04-075-01.png
 96 | d04-075-08.png
 97 | d04-081-01.png
 98 | d04-081-00.png
 99 | d04-081-08.png
100 | d04-081-03.png
101 | d04-081-06.png
102 | d04-081-04.png
103 | d04-081-05.png
104 | d04-081-02.png
105 | d04-081-07.png
106 | d04-117-00.png
107 | d04-117-05.png
108 | d04-117-03.png
109 | d04-117-04.png
110 | d04-117-06.png
111 | d04-117-01.png
112 | d04-117-07.png
113 | d04-117-02.png
114 | d04-121-03.png
115 | d04-121-06.png
116 | d04-121-05.png
117 | d04-121-00.png
118 | d04-121-02.png
119 | d04-121-04.png
120 | d04-121-01.png
121 | d06-000-07.png
122 | d06-000-00.png
123 | d06-000-03.png
124 | d06-000-04.png
125 | d06-000-09.png
126 | d06-000-02.png
127 | d06-000-08.png
128 | d06-000-01.png
129 | d06-000-06.png
130 | d06-000-05.png
131 | d06-015-03.png
132 | d06-015-00.png
133 | d06-015-01.png
134 | d06-015-05.png
135 | d06-015-02.png
136 | d06-015-04.png
137 | d06-030-03.png
138 | d06-030-04.png
139 | d06-030-01.png
140 | d06-030-00.png
141 | d06-030-02.png
142 | d06-030-05.png
143 | d06-050-03.png
144 | d06-050-00.png
145 | d06-050-06.png
146 | d06-050-05.png
147 | d06-050-02.png
148 | d06-050-01.png
149 | d06-050-04.png
150 | d06-082-02.png
151 | d06-082-01.png
152 | d06-082-00.png
153 | d06-082-04.png
154 | d06-082-05.png
155 | d06-082-03.png
156 | d06-091-08.png
157 | d06-091-00.png
158 | d06-091-03.png
159 | d06-091-05.png
160 | d06-091-07.png
161 | d06-091-01.png
162 | d06-091-04.png
163 | d06-091-02.png
164 | d06-091-10.png
165 | d06-091-06.png
166 | d06-091-09.png
167 | e01-014-04.png
168 | e01-014-00.png
169 | e01-014-05.png
170 | e01-014-06.png
171 | e01-014-02.png
172 | e01-014-01.png
173 | e01-014-03.png
174 | e01-018-04.png
175 | e01-018-02.png
176 | e01-018-07.png
177 | e01-018-00.png
178 | e01-018-05.png
179 | e01-018-01.png
180 | e01-018-08.png
181 | e01-018-06.png
182 | e01-018-03.png
183 | e06-006-04.png
184 | e06-006-07.png
185 | e06-006-01.png
186 | e06-006-08.png
187 | e06-006-09.png
188 | e06-006-03.png
189 | e06-006-06.png
190 | e06-006-10.png
191 | e06-006-02.png
192 | e06-006-05.png
193 | e06-006-00.png
194 | e06-037-08.png
195 | e06-037-02.png
196 | e06-037-06.png
197 | e06-037-00.png
198 | e06-037-09.png
199 | e06-037-04.png
200 | e06-037-05.png
201 | e06-037-10.png
202 | e06-037-03.png
203 | e06-037-07.png
204 | e06-037-01.png
205 | e06-041-00.png
206 | e06-041-02.png
207 | e06-041-04.png
208 | e06-041-05.png
209 | e06-041-06.png
210 | e06-041-01.png
211 | e06-041-07.png
212 | e06-041-08.png
213 | e06-041-03.png
214 | e06-041-09.png
215 | f04-046-03.png
216 | f04-046-05.png
217 | f04-046-02.png
218 | f04-046-01.png
219 | f04-046-04.png
220 | f04-046-00.png
221 | f04-049-05.png
222 | f04-049-03.png
223 | f04-049-04.png
224 | f04-049-02.png
225 | f04-049-00.png
226 | f04-049-01.png
227 | f04-049-06.png
228 | f04-093-03.png
229 | f04-093-06.png
230 | f04-093-05.png
231 | f04-093-08.png
232 | f04-093-09.png
233 | f04-093-04.png
234 | f04-093-02.png
235 | f04-093-00.png
236 | f04-093-07.png
237 | f04-093-01.png
238 | f04-096-08.png
239 | f04-096-07.png
240 | f04-096-09.png
241 | f04-096-10.png
242 | f04-096-05.png
243 | f04-096-04.png
244 | f04-096-03.png
245 | f04-096-01.png
246 | f04-096-00.png
247 | f04-096-06.png
248 | f04-096-02.png
249 | f04-100-03.png
250 | f04-100-00.png
251 | f04-100-10.png
252 | f04-100-02.png
253 | f04-100-08.png
254 | f04-100-01.png
255 | f04-100-04.png
256 | f04-100-06.png
257 | f04-100-07.png
258 | f04-100-05.png
259 | f04-100-09.png
260 | f07-028b-06.png
261 | f07-028b-00.png
262 | f07-028b-04.png
263 | f07-028b-05.png
264 | f07-028b-09.png
265 | f07-028b-01.png
266 | f07-028b-03.png
267 | f07-028b-10.png
268 | f07-028b-02.png
269 | f07-028b-07.png
270 | f07-028b-11.png
271 | f07-028b-08.png
272 | f07-032b-09.png
273 | f07-032b-04.png
274 | f07-032b-06.png
275 | f07-032b-08.png
276 | f07-032b-00.png
277 | f07-032b-05.png
278 | f07-032b-10.png
279 | f07-032b-01.png
280 | f07-032b-03.png
281 | f07-032b-11.png
282 | f07-032b-02.png
283 | f07-032b-07.png
284 | f07-036-06.png
285 | f07-036-08.png
286 | f07-036-01.png
287 | f07-036-07.png
288 | f07-036-02.png
289 | f07-036-03.png
290 | f07-036-00.png
291 | f07-036-04.png
292 | f07-036-05.png
293 | f07-039b-06.png
294 | f07-039b-08.png
295 | f07-039b-04.png
296 | f07-039b-07.png
297 | f07-039b-05.png
298 | f07-039b-03.png
299 | f07-039b-00.png
300 | f07-039b-01.png
301 | f07-039b-02.png
302 | f07-042b-02.png
303 | f07-042b-05.png
304 | f07-042b-00.png
305 | f07-042b-04.png
306 | f07-042b-03.png
307 | f07-042b-06.png
308 | f07-042b-07.png
309 | f07-042b-01.png
310 | f07-046b-06.png
311 | f07-046b-08.png
312 | f07-046b-07.png
313 | f07-046b-04.png
314 | f07-046b-05.png
315 | f07-046b-03.png
316 | f07-046b-01.png
317 | f07-046b-02.png
318 | f07-046b-00.png
319 | g01-083-08.png
320 | g01-083-02.png
321 | g01-083-10.png
322 | g01-083-07.png
323 | g01-083-00.png
324 | g01-083-04.png
325 | g01-083-09.png
326 | g01-083-01.png
327 | g01-083-03.png
328 | g01-083-05.png
329 | g01-083-06.png
330 | g03-016-02.png
331 | g03-016-06.png
332 | g03-016-08.png
333 | g03-016-00.png
334 | g03-016-01.png
335 | g03-016-05.png
336 | g03-016-03.png
337 | g03-016-04.png
338 | g03-016-07.png
339 | g03-058-05.png
340 | g03-058-07.png
341 | g03-058-04.png
342 | g03-058-06.png
343 | g03-058-01.png
344 | g03-058-00.png
345 | g03-058-03.png
346 | g03-058-02.png
347 | g04-036-05.png
348 | g04-036-06.png
349 | g04-036-07.png
350 | g04-036-04.png
351 | g04-036-03.png
352 | g04-036-00.png
353 | g04-036-02.png
354 | g04-036-01.png
355 | g04-039-05.png
356 | g04-039-01.png
357 | g04-039-03.png
358 | g04-039-04.png
359 | g04-039-00.png
360 | g04-039-06.png
361 | g04-039-07.png
362 | g04-039-02.png
363 | g04-055-01.png
364 | g04-055-06.png
365 | g04-055-08.png
366 | g04-055-02.png
367 | g04-055-00.png
368 | g04-055-05.png
369 | g04-055-04.png
370 | g04-055-07.png
371 | g04-055-03.png
372 | g04-055-09.png
373 | g04-060-04.png
374 | g04-060-05.png
375 | g04-060-03.png
376 | g04-060-00.png
377 | g04-060-02.png
378 | g04-060-06.png
379 | g04-060-01.png
380 | g04-063-02.png
381 | g04-063-09.png
382 | g04-063-06.png
383 | g04-063-07.png
384 | g04-063-00.png
385 | g04-063-01.png
386 | g04-063-08.png
387 | g04-063-03.png
388 | g04-063-05.png
389 | g04-063-04.png
390 | h01-004-03.png
391 | h01-004-05.png
392 | h01-004-01.png
393 | h01-004-04.png
394 | h01-004-02.png
395 | h01-004-00.png
396 | m01-090-04.png
397 | m01-090-05.png
398 | m01-090-07.png
399 | m01-090-03.png
400 | m01-090-00.png
401 | m01-090-02.png
402 | m01-090-06.png
403 | m01-090-01.png
404 | m01-115-01.png
405 | m01-115-02.png
406 | m01-115-00.png
407 | m01-115-03.png
408 | m01-115-05.png
409 | m01-115-06.png
410 | m01-115-04.png
411 | m01-115-08.png
412 | m01-115-07.png
413 | m01-160-05.png
414 | m01-160-04.png
415 | m01-160-02.png
416 | m01-160-00.png
417 | m01-160-06.png
418 | m01-160-01.png
419 | m01-160-03.png
420 | m02-052-04.png
421 | m02-052-03.png
422 | m02-052-06.png
423 | m02-052-05.png
424 | m02-052-02.png
425 | m02-052-01.png
426 | m02-052-00.png
427 | m02-075-04.png
428 | m02-075-02.png
429 | m02-075-06.png
430 | m02-075-05.png
431 | m02-075-00.png
432 | m02-075-01.png
433 | m02-075-03.png
434 | m02-075-07.png
435 | m02-083-09.png
436 | m02-083-01.png
437 | m02-083-04.png
438 | m02-083-08.png
439 | m02-083-11.png
440 | m02-083-07.png
441 | m02-083-02.png
442 | m02-083-06.png
443 | m02-083-03.png
444 | m02-083-10.png
445 | m02-083-05.png
446 | m02-083-00.png
447 | m02-090-02.png
448 | m02-090-00.png
449 | m02-090-08.png
450 | m02-090-07.png
451 | m02-090-04.png
452 | m02-090-06.png
453 | m02-090-10.png
454 | m02-090-05.png
455 | m02-090-09.png
456 | m02-090-03.png
457 | m02-090-01.png
458 | m02-102-06.png
459 | m02-102-07.png
460 | m02-102-04.png
461 | m02-102-08.png
462 | m02-102-02.png
463 | m02-102-09.png
464 | m02-102-05.png
465 | m02-102-00.png
466 | m02-102-10.png
467 | m02-102-01.png
468 | m02-102-03.png
469 | m02-106-00.png
470 | m02-106-02.png
471 | m02-106-06.png
472 | m02-106-09.png
473 | m02-106-04.png
474 | m02-106-01.png
475 | m02-106-07.png
476 | m02-106-08.png
477 | m02-106-05.png
478 | m02-106-03.png
479 | m03-062-05.png
480 | m03-062-01.png
481 | m03-062-06.png
482 | m03-062-00.png
483 | m03-062-03.png
484 | m03-062-08.png
485 | m03-062-07.png
486 | m03-062-04.png
487 | m03-062-02.png
488 | m03-095-00.png
489 | m03-095-06.png
490 | m03-095-05.png
491 | m03-095-07.png
492 | m03-095-01.png
493 | m03-095-03.png
494 | m03-095-08.png
495 | m03-095-02.png
496 | m03-095-04.png
497 | m04-000-04.png
498 | m04-000-06.png
499 | m04-000-03.png
500 | m04-000-05.png
501 | m04-000-00.png
502 | m04-000-07.png
503 | m04-000-01.png
504 | m04-000-02.png
505 | m04-007-08.png
506 | m04-007-03.png
507 | m04-007-02.png
508 | m04-007-00.png
509 | m04-007-06.png
510 | m04-007-05.png
511 | m04-007-09.png
512 | m04-007-01.png
513 | m04-007-04.png
514 | m04-007-07.png
515 | m04-012-09.png
516 | m04-012-07.png
517 | m04-012-04.png
518 | m04-012-06.png
519 | m04-012-08.png
520 | m04-012-00.png
521 | m04-012-01.png
522 | m04-012-02.png
523 | m04-012-03.png
524 | m04-012-05.png
525 | m04-019-01.png
526 | m04-019-07.png
527 | m04-019-00.png
528 | m04-019-06.png
529 | m04-019-05.png
530 | m04-019-04.png
531 | m04-019-03.png
532 | m04-019-02.png
533 | m04-024-06.png
534 | m04-024-01.png
535 | m04-024-07.png
536 | m04-024-02.png
537 | m04-024-00.png
538 | m04-024-03.png
539 | m04-024-05.png
540 | m04-024-04.png
541 | m04-145-04.png
542 | m04-145-00.png
543 | m04-145-05.png
544 | m04-145-01.png
545 | m04-145-02.png
546 | m04-145-03.png
547 | m04-145-06.png
548 | m04-145-07.png
549 | m04-152-03.png
550 | m04-152-01.png
551 | m04-152-06.png
552 | m04-152-00.png
553 | m04-152-02.png
554 | m04-152-05.png
555 | m04-152-07.png
556 | m04-152-04.png
557 | m04-164-07.png
558 | m04-164-00.png
559 | m04-164-04.png
560 | m04-164-01.png
561 | m04-164-03.png
562 | m04-164-02.png
563 | m04-164-05.png
564 | m04-164-06.png
565 | m04-180-02.png
566 | m04-180-03.png
567 | m04-180-06.png
568 | m04-180-04.png
569 | m04-180-01.png
570 | m04-180-00.png
571 | m04-180-05.png
572 | m04-180-07.png
573 | m04-190-08.png
574 | m04-190-05.png
575 | m04-190-02.png
576 | m04-190-04.png
577 | m04-190-07.png
578 | m04-190-00.png
579 | m04-190-06.png
580 | m04-190-01.png
581 | m04-190-03.png
582 | m04-190-09.png
583 | m04-200-03.png
584 | m04-200-06.png
585 | m04-200-02.png
586 | m04-200-01.png
587 | m04-200-07.png
588 | m04-200-04.png
589 | m04-200-05.png
590 | m04-200-00.png
591 | m04-209-01.png
592 | m04-209-02.png
593 | m04-209-06.png
594 | m04-209-04.png
595 | m04-209-05.png
596 | m04-209-00.png
597 | m04-209-03.png
598 | m04-216-02.png
599 | m04-216-01.png
600 | m04-216-06.png
601 | m04-216-05.png
602 | m04-216-04.png
603 | m04-216-03.png
604 | m04-216-00.png
605 | m04-222-05.png
606 | m04-222-07.png
607 | m04-222-03.png
608 | m04-222-00.png
609 | m04-222-01.png
610 | m04-222-04.png
611 | m04-222-06.png
612 | m04-222-02.png
613 | m04-251-00.png
614 | m04-251-01.png
615 | m04-251-02.png
616 | m06-019-04.png
617 | m06-019-01.png
618 | m06-019-03.png
619 | m06-019-05.png
620 | m06-019-06.png
621 | m06-019-02.png
622 | m06-019-00.png
623 | m06-019-07.png
624 | n02-000-07.png
625 | n02-000-01.png
626 | n02-000-04.png
627 | n02-000-06.png
628 | n02-000-02.png
629 | n02-000-05.png
630 | n02-000-00.png
631 | n02-000-03.png
632 | n02-016-01.png
633 | n02-016-06.png
634 | n02-016-03.png
635 | n02-016-07.png
636 | n02-016-05.png
637 | n02-016-02.png
638 | n02-016-00.png
639 | n02-016-04.png
640 | n02-082a-00.png
641 | n02-082a-02.png
642 | n02-082a-05.png
643 | n02-082a-04.png
644 | n02-082a-06.png
645 | n02-082a-01.png
646 | n02-082a-03.png
647 | n02-098-03.png
648 | n02-098-06.png
649 | n02-098-04.png
650 | n02-098-05.png
651 | n02-098-00.png
652 | n02-098-01.png
653 | n02-098-07.png
654 | n02-098-02.png
655 | n03-079-02.png
656 | n03-079-06.png
657 | n03-079-05.png
658 | n03-079-00.png
659 | n03-079-01.png
660 | n03-079-04.png
661 | n03-079-03.png
662 | n03-082-04.png
663 | n03-082-02.png
664 | n03-082-05.png
665 | n03-082-03.png
666 | n03-082-01.png
667 | n03-082-00.png
668 | n03-082-06.png
669 | n03-082-07.png
670 | n03-091-02.png
671 | n03-091-00.png
672 | n03-091-01.png
673 | n03-091-04.png
674 | n03-091-08.png
675 | n03-091-03.png
676 | n03-091-07.png
677 | n03-091-06.png
678 | n03-091-05.png
679 | n04-015-07.png
680 | n04-015-01.png
681 | n04-015-03.png
682 | n04-015-05.png
683 | n04-015-02.png
684 | n04-015-04.png
685 | n04-015-00.png
686 | n04-015-06.png
687 | n04-022-04.png
688 | n04-022-07.png
689 | n04-022-05.png
690 | n04-022-06.png
691 | n04-022-03.png
692 | n04-022-00.png
693 | n04-022-01.png
694 | n04-022-02.png
695 | n04-031-00.png
696 | n04-031-07.png
697 | n04-031-03.png
698 | n04-031-01.png
699 | n04-031-06.png
700 | n04-031-05.png
701 | n04-031-02.png
702 | n04-031-04.png
703 | n04-039-03.png
704 | n04-039-02.png
705 | n04-039-04.png
706 | n04-039-01.png
707 | n04-039-05.png
708 | n04-039-00.png
709 | n04-044-10.png
710 | n04-044-05.png
711 | n04-044-02.png
712 | n04-044-07.png
713 | n04-044-04.png
714 | n04-044-00.png
715 | n04-044-09.png
716 | n04-044-03.png
717 | n04-044-08.png
718 | n04-044-06.png
719 | n04-044-01.png
720 | n04-171-00.png
721 | n04-171-04.png
722 | n04-171-06.png
723 | n04-171-05.png
724 | n04-171-02.png
725 | n04-171-03.png
726 | n04-171-01.png
727 | n04-183-00.png
728 | n04-183-05.png
729 | n04-183-03.png
730 | n04-183-01.png
731 | n04-183-04.png
732 | n04-183-06.png
733 | n04-183-02.png
734 | n04-190-03.png
735 | n04-190-04.png
736 | n04-190-01.png
737 | n04-190-00.png
738 | n04-190-02.png
739 | n04-190-05.png
740 | n04-190-06.png
741 | n04-195-03.png
742 | n04-195-01.png
743 | n04-195-04.png
744 | n04-195-02.png
745 | n04-195-05.png
746 | n04-195-00.png
747 | n06-148-00.png
748 | n06-148-01.png
749 | n06-148-08.png
750 | n06-148-07.png
751 | n06-148-02.png
752 | n06-148-05.png
753 | n06-148-06.png
754 | n06-148-04.png
755 | n06-148-03.png
756 | n06-156-04.png
757 | n06-156-00.png
758 | n06-156-05.png
759 | n06-156-02.png
760 | n06-156-06.png
761 | n06-156-01.png
762 | n06-156-03.png
763 | n06-156-07.png
764 | n06-163-03.png
765 | n06-163-08.png
766 | n06-163-09.png
767 | n06-163-01.png
768 | n06-163-05.png
769 | n06-163-02.png
770 | n06-163-04.png
771 | n06-163-00.png
772 | n06-163-07.png
773 | n06-163-06.png
774 | n06-169-08.png
775 | n06-169-01.png
776 | n06-169-02.png
777 | n06-169-00.png
778 | n06-169-04.png
779 | n06-169-05.png
780 | n06-169-06.png
781 | n06-169-03.png
782 | n06-169-07.png
783 | n06-175-05.png
784 | n06-175-09.png
785 | n06-175-08.png
786 | n06-175-07.png
787 | n06-175-04.png
788 | n06-175-02.png
789 | n06-175-03.png
790 | n06-175-01.png
791 | n06-175-06.png
792 | n06-175-00.png
793 | n06-182-06.png
794 | n06-182-00.png
795 | n06-182-01.png
796 | n06-182-03.png
797 | n06-182-04.png
798 | n06-182-02.png
799 | n06-182-07.png
800 | n06-182-05.png
801 | n06-186-01.png
802 | n06-186-05.png
803 | n06-186-00.png
804 | n06-186-07.png
805 | n06-186-04.png
806 | n06-186-02.png
807 | n06-186-06.png
808 | n06-186-03.png
809 | n06-194-05.png
810 | n06-194-04.png
811 | n06-194-07.png
812 | n06-194-00.png
813 | n06-194-03.png
814 | n06-194-01.png
815 | n06-194-02.png
816 | n06-194-06.png
817 | n06-201-01.png
818 | n06-201-03.png
819 | n06-201-05.png
820 | n06-201-04.png
821 | n06-201-06.png
822 | n06-201-00.png
823 | n06-201-02.png
824 | n06-201-07.png
825 | p02-017-06.png
826 | p02-017-01.png
827 | p02-017-00.png
828 | p02-017-03.png
829 | p02-017-04.png
830 | p02-017-05.png
831 | p02-017-02.png
832 | p02-017-07.png
833 | p02-121-02.png
834 | p02-121-04.png
835 | p02-121-05.png
836 | p02-121-03.png
837 | p02-121-01.png
838 | p02-121-00.png
839 | p02-127-01.png
840 | p02-127-03.png
841 | p02-127-02.png
842 | p02-127-04.png
843 | p02-127-05.png
844 | p02-127-00.png
845 | p02-127-06.png
846 | p02-131-05.png
847 | p02-131-01.png
848 | p02-131-07.png
849 | p02-131-06.png
850 | p02-131-03.png
851 | p02-131-02.png
852 | p02-131-00.png
853 | p02-131-04.png
854 | p02-155-01.png
855 | p02-155-00.png
856 | p03-004-06.png
857 | p03-004-00.png
858 | p03-004-04.png
859 | p03-004-01.png
860 | p03-004-09.png
861 | p03-004-07.png
862 | p03-004-03.png
863 | p03-004-05.png
864 | p03-004-10.png
865 | p03-004-08.png
866 | p03-004-02.png
867 | p03-004-11.png
868 | p03-009-03.png
869 | p03-009-01.png
870 | p03-009-04.png
871 | p03-009-05.png
872 | p03-009-00.png
873 | p03-009-02.png
874 | p03-009-06.png
875 | p03-057-07.png
876 | p03-057-04.png
877 | p03-057-08.png
878 | p03-057-02.png
879 | p03-057-09.png
880 | p03-057-03.png
881 | p03-057-05.png
882 | p03-057-00.png
883 | p03-057-06.png
884 | p03-057-01.png
885 | p03-069-01.png
886 | p03-069-00.png
887 | p03-069-02.png
888 | p03-069-05.png
889 | p03-069-06.png
890 | p03-069-04.png
891 | p03-069-03.png
892 | p03-087-06.png
893 | p03-087-03.png
894 | p03-087-10.png
895 | p03-087-08.png
896 | p03-087-01.png
897 | p03-087-00.png
898 | p03-087-07.png
899 | p03-087-05.png
900 | p03-087-04.png
901 | p03-087-09.png
902 | p03-087-02.png
903 | p03-096-08.png
904 | p03-096-06.png
905 | p03-096-09.png
906 | p03-096-02.png
907 | p03-096-04.png
908 | p03-096-05.png
909 | p03-096-03.png
910 | p03-096-00.png
911 | p03-096-01.png
912 | p03-096-07.png
913 | p03-103-06.png
914 | p03-103-05.png
915 | p03-103-00.png
916 | p03-103-03.png
917 | p03-103-02.png
918 | p03-103-01.png
919 | p03-103-04.png
920 | p03-103-07.png
921 | p03-112-02.png
922 | p03-112-00.png
923 | p03-112-05.png
924 | p03-112-06.png
925 | p03-112-07.png
926 | p03-112-01.png
927 | p03-112-08.png
928 | p03-112-03.png
929 | p03-112-04.png
930 | p03-112-09.png
931 | p03-121-04.png
932 | p03-121-01.png
933 | p03-121-03.png
934 | p03-121-06.png
935 | p03-121-02.png
936 | p03-121-00.png
937 | p03-121-05.png
938 | p03-135-06.png
939 | p03-135-00.png
940 | p03-135-05.png
941 | p03-135-03.png
942 | p03-135-04.png
943 | p03-135-02.png
944 | p03-135-01.png
945 | p03-142-02.png
946 | p03-142-03.png
947 | p03-142-06.png
948 | p03-142-05.png
949 | p03-142-00.png
950 | p03-142-01.png
951 | p03-142-04.png
952 | p03-151-06.png
953 | p03-151-00.png
954 | p03-151-07.png
955 | p03-151-05.png
956 | p03-151-03.png
957 | p03-151-02.png
958 | p03-151-04.png
959 | p03-151-01.png
960 | p03-158-05.png
961 | p03-158-02.png
962 | p03-158-03.png
963 | p03-158-06.png
964 | p03-158-01.png
965 | p03-158-07.png
966 | p03-158-00.png
967 | p03-158-04.png
968 | p03-185-02.png
969 | p03-185-06.png
970 | p03-185-07.png
971 | p03-185-03.png
972 | p03-185-00.png
973 | p03-185-05.png
974 | p03-185-08.png
975 | p03-185-04.png
976 | p03-185-01.png
977 | 


--------------------------------------------------------------------------------
/ich17/ich.gin:
--------------------------------------------------------------------------------
 1 | port = 12345
 2 | manualSeed = 1111
 3 | dist = 'DDP'
 4 | 
 5 | n_channels = 3
 6 | o_classes = 92
 7 | 
 8 | GradCheck = 1
 9 | 
10 | # Parameters for get_images (ds_load.py):
11 | # ==============================================================================
12 | get_images.max_w = 1000
13 | get_images.max_h = 1400
14 | get_images.nch = %n_channels
15 | 
16 | # Parameters for SameTrCollate (ds_load.py):
17 | # ==============================================================================
18 | SameTrCollate.prjAug = True
19 | SameTrCollate.prjVal = 125
20 | 
21 | # Parameters for train (train.py):
22 | # ==============================================================================
23 | train.experiment_name = 'ich_gin_test_'
24 | 
25 | train.AMP = False
26 | train.WdB = False
27 | train.wdbprj = 'pt-fpr-ich17'
28 | train.train_data_list = 'ich17/train'
29 | train.train_data_path = '/mnt/ich17/Train-B/'
30 | train.test_data_list  = 'ich17/test'
31 | train.test_data_path  = '/mnt/ich17/Train-B/'
32 | train.train_batch_size = 1
33 | train.val_batch_size   = 1
34 | train.workers = 4
35 | train.lr = 0.01
36 | train.continue_model = ''
37 | train.valInterval = 500
38 | train.num_iter = 0
39 | 
40 | 
41 | # Parameters for OrigamiNet (cnv_model.py):
42 | # ==============================================================================
43 | OrigamiNet.n_channels = %n_channels
44 | OrigamiNet.o_classes  = %o_classes
45 | OrigamiNet.fup = 33
46 | OrigamiNet.GradCheck = %GradCheck
47 | 
48 | OrigamiNet.nlyrs = 12
49 | OrigamiNet.wmul = 1.0
50 | 
51 | OrigamiNet.lszs = {0:  128,
52 |             2:  256,
53 |             4:  512,
54 |             11: 256  }
55 | 
56 | ##resize structure##
57 | MaxPool2d.kernel_size = (2,2)
58 | 
59 | Upsample.mode = 'bilinear'
60 | Upsample.align_corners = True
61 | s1/Upsample.size = (3500,31)
62 | s2/Upsample.size = (7200,15)
63 | 
64 | OrigamiNet.lreszs = {
65 |                   0: @MaxPool2d(),
66 |                   2: @MaxPool2d(),
67 |                   
68 |                   4: @PadPool(),
69 |                   6: @PadPool(),
70 |                   8: @PadPool(),
71 |                   
72 |                   10: @s1/Upsample(),
73 |                   11: @s2/Upsample()
74 |             }
75 | 
76 | # Parameters for InitBlock (cnv_model.py):
77 | # ==============================================================================
78 | InitBlock.n_channels = %n_channels
79 | 
80 | # Parameters for GateBlock (cnv_model.py):
81 | # ==============================================================================
82 | GateBlock.GradCheck = %GradCheck


--------------------------------------------------------------------------------
/ich17/ich.py:
--------------------------------------------------------------------------------
 1 | import xml.etree.ElementTree as ET
 2 | from xml.sax.saxutils import unescape
 3 | import glob
 4 | import sys,os
 5 | import scipy.misc
 6 | from PIL import Image
 7 | import os.path as osp
 8 | import unicodedata
 9 | 
10 | base  = sys.argv[1]
11 | 
12 | xml    = osp.join(base,'page/')
13 | pages  = glob.glob(xml+'*.xml')
14 | 
15 | rm = ["§","æ","­","|","‰","#","+","[","]","œ","̃","‒","*","□","°","†","‹","›","ο","—","£","τ","ν","‡","ψ","ι","α","κ","ω","η","℔","	","χ","ξ","₤","ε","π","~","μ","¬","Ζ","λ","Τ","Γ","І","̸","∫","·",">","♂","✓","Œ","♀","$","∆","ø","ρ","∇"]
16 | print(len(pages))
17 | a=[]
18 | for i in range(len(pages)):
19 |     rt = ET.parse(pages[i]).getroot()
20 |     un = rt.find(".//{http://schema.primaresearch.org/PAGE/gts/pagecontent/2013-07-15}Unicode")
21 |     txt = unescape(un.text, {"&apos;": "'", "&quot;": '"'})
22 |     txt = unicodedata.normalize('NFKD', txt)
23 |     txt = txt.translate({ord(i): None for i in rm})
24 |     
25 |     a.append(txt)
26 |     open(osp.join(base,pages[i][-10:-4]+'.txt'),'w+').write(txt.strip())


--------------------------------------------------------------------------------
/ich17/ich.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | 
 3 | dst=$1
 4 | 
 5 | wget --content-disposition  https://zenodo.org/record/439811/files/Train-B_batch{1,2}.tbz2?download=1
 6 | 
 7 | mkdir -p $dst
 8 | 
 9 | cat Train-B_batch{1,2}.tbz2 | tar -jxvf - -i --strip-components=2 -C $dst
10 | 
11 | python `dirname "$0"`/ich.py $dst


--------------------------------------------------------------------------------
/ich17/test:
--------------------------------------------------------------------------------
   1 | 019001.jpg
   2 | 019002.jpg
   3 | 019003.jpg
   4 | 019004.jpg
   5 | 019005.jpg
   6 | 019006.jpg
   7 | 019007.jpg
   8 | 019008.jpg
   9 | 019009.jpg
  10 | 019010.jpg
  11 | 019011.jpg
  12 | 019012.jpg
  13 | 019013.jpg
  14 | 019014.jpg
  15 | 019015.jpg
  16 | 019016.jpg
  17 | 019017.jpg
  18 | 019018.jpg
  19 | 019019.jpg
  20 | 019020.jpg
  21 | 019021.jpg
  22 | 019022.jpg
  23 | 019023.jpg
  24 | 019024.jpg
  25 | 019025.jpg
  26 | 019026.jpg
  27 | 019027.jpg
  28 | 019028.jpg
  29 | 019029.jpg
  30 | 019030.jpg
  31 | 019031.jpg
  32 | 019032.jpg
  33 | 019033.jpg
  34 | 019034.jpg
  35 | 019035.jpg
  36 | 019036.jpg
  37 | 019037.jpg
  38 | 019038.jpg
  39 | 019039.jpg
  40 | 019040.jpg
  41 | 019041.jpg
  42 | 019042.jpg
  43 | 019043.jpg
  44 | 019044.jpg
  45 | 019045.jpg
  46 | 019046.jpg
  47 | 019047.jpg
  48 | 019048.jpg
  49 | 019049.jpg
  50 | 019050.jpg
  51 | 019051.jpg
  52 | 019052.jpg
  53 | 019053.jpg
  54 | 019054.jpg
  55 | 019055.jpg
  56 | 019056.jpg
  57 | 019057.jpg
  58 | 019058.jpg
  59 | 019059.jpg
  60 | 019060.jpg
  61 | 019061.jpg
  62 | 019062.jpg
  63 | 019063.jpg
  64 | 019064.jpg
  65 | 019065.jpg
  66 | 019066.jpg
  67 | 019067.jpg
  68 | 019068.jpg
  69 | 019069.jpg
  70 | 019070.jpg
  71 | 019071.jpg
  72 | 019072.jpg
  73 | 019073.jpg
  74 | 019074.jpg
  75 | 019075.jpg
  76 | 019076.jpg
  77 | 019077.jpg
  78 | 019078.jpg
  79 | 019079.jpg
  80 | 019080.jpg
  81 | 019081.jpg
  82 | 019082.jpg
  83 | 019083.jpg
  84 | 019084.jpg
  85 | 019085.jpg
  86 | 019086.jpg
  87 | 019087.jpg
  88 | 019088.jpg
  89 | 019089.jpg
  90 | 019090.jpg
  91 | 019091.jpg
  92 | 019092.jpg
  93 | 019093.jpg
  94 | 019094.jpg
  95 | 019095.jpg
  96 | 019096.jpg
  97 | 019097.jpg
  98 | 019098.jpg
  99 | 019099.jpg
 100 | 019100.jpg
 101 | 019101.jpg
 102 | 019102.jpg
 103 | 019103.jpg
 104 | 019104.jpg
 105 | 019105.jpg
 106 | 019106.jpg
 107 | 019107.jpg
 108 | 019108.jpg
 109 | 019109.jpg
 110 | 019110.jpg
 111 | 019111.jpg
 112 | 019112.jpg
 113 | 019113.jpg
 114 | 019114.jpg
 115 | 019115.jpg
 116 | 019116.jpg
 117 | 019117.jpg
 118 | 019118.jpg
 119 | 019119.jpg
 120 | 019120.jpg
 121 | 019121.jpg
 122 | 019122.jpg
 123 | 019123.jpg
 124 | 019124.jpg
 125 | 019125.jpg
 126 | 019126.jpg
 127 | 019127.jpg
 128 | 019128.jpg
 129 | 019129.jpg
 130 | 019130.jpg
 131 | 019131.jpg
 132 | 019132.jpg
 133 | 019133.jpg
 134 | 019134.jpg
 135 | 019135.jpg
 136 | 019136.jpg
 137 | 019137.jpg
 138 | 019138.jpg
 139 | 019139.jpg
 140 | 019140.jpg
 141 | 019141.jpg
 142 | 019142.jpg
 143 | 019143.jpg
 144 | 019144.jpg
 145 | 019145.jpg
 146 | 019146.jpg
 147 | 019147.jpg
 148 | 019148.jpg
 149 | 019149.jpg
 150 | 019150.jpg
 151 | 019151.jpg
 152 | 019152.jpg
 153 | 019153.jpg
 154 | 019154.jpg
 155 | 019155.jpg
 156 | 019156.jpg
 157 | 019157.jpg
 158 | 019158.jpg
 159 | 019159.jpg
 160 | 019160.jpg
 161 | 019161.jpg
 162 | 019162.jpg
 163 | 019163.jpg
 164 | 019164.jpg
 165 | 019165.jpg
 166 | 019166.jpg
 167 | 019167.jpg
 168 | 019168.jpg
 169 | 019169.jpg
 170 | 019170.jpg
 171 | 019171.jpg
 172 | 019172.jpg
 173 | 019173.jpg
 174 | 019174.jpg
 175 | 019175.jpg
 176 | 019176.jpg
 177 | 019177.jpg
 178 | 019178.jpg
 179 | 019179.jpg
 180 | 019180.jpg
 181 | 019181.jpg
 182 | 019182.jpg
 183 | 019183.jpg
 184 | 019184.jpg
 185 | 019185.jpg
 186 | 019186.jpg
 187 | 019187.jpg
 188 | 019188.jpg
 189 | 019189.jpg
 190 | 019190.jpg
 191 | 019191.jpg
 192 | 019192.jpg
 193 | 019193.jpg
 194 | 019194.jpg
 195 | 019195.jpg
 196 | 019196.jpg
 197 | 019197.jpg
 198 | 019198.jpg
 199 | 019199.jpg
 200 | 019200.jpg
 201 | 019201.jpg
 202 | 019202.jpg
 203 | 019203.jpg
 204 | 019204.jpg
 205 | 019205.jpg
 206 | 019206.jpg
 207 | 019207.jpg
 208 | 019208.jpg
 209 | 019209.jpg
 210 | 019210.jpg
 211 | 019211.jpg
 212 | 019212.jpg
 213 | 019213.jpg
 214 | 019214.jpg
 215 | 019215.jpg
 216 | 019216.jpg
 217 | 019217.jpg
 218 | 019218.jpg
 219 | 019219.jpg
 220 | 019220.jpg
 221 | 019221.jpg
 222 | 019222.jpg
 223 | 019223.jpg
 224 | 019224.jpg
 225 | 019225.jpg
 226 | 019226.jpg
 227 | 019227.jpg
 228 | 019228.jpg
 229 | 019229.jpg
 230 | 019230.jpg
 231 | 019231.jpg
 232 | 019232.jpg
 233 | 019233.jpg
 234 | 019234.jpg
 235 | 019235.jpg
 236 | 019236.jpg
 237 | 019237.jpg
 238 | 019238.jpg
 239 | 019239.jpg
 240 | 019240.jpg
 241 | 019241.jpg
 242 | 019242.jpg
 243 | 019243.jpg
 244 | 019244.jpg
 245 | 019245.jpg
 246 | 019246.jpg
 247 | 019247.jpg
 248 | 019248.jpg
 249 | 019249.jpg
 250 | 019250.jpg
 251 | 019251.jpg
 252 | 019252.jpg
 253 | 019253.jpg
 254 | 019254.jpg
 255 | 019255.jpg
 256 | 019256.jpg
 257 | 019257.jpg
 258 | 019258.jpg
 259 | 019259.jpg
 260 | 019260.jpg
 261 | 019261.jpg
 262 | 019262.jpg
 263 | 019263.jpg
 264 | 019264.jpg
 265 | 019265.jpg
 266 | 019266.jpg
 267 | 019267.jpg
 268 | 019268.jpg
 269 | 019269.jpg
 270 | 019270.jpg
 271 | 019271.jpg
 272 | 019272.jpg
 273 | 019273.jpg
 274 | 019274.jpg
 275 | 019275.jpg
 276 | 019276.jpg
 277 | 019277.jpg
 278 | 019278.jpg
 279 | 019279.jpg
 280 | 019280.jpg
 281 | 019281.jpg
 282 | 019282.jpg
 283 | 019283.jpg
 284 | 019284.jpg
 285 | 019285.jpg
 286 | 019286.jpg
 287 | 019287.jpg
 288 | 019288.jpg
 289 | 019289.jpg
 290 | 019290.jpg
 291 | 019291.jpg
 292 | 019292.jpg
 293 | 019293.jpg
 294 | 019294.jpg
 295 | 019295.jpg
 296 | 019296.jpg
 297 | 019297.jpg
 298 | 019298.jpg
 299 | 019299.jpg
 300 | 019300.jpg
 301 | 019301.jpg
 302 | 019302.jpg
 303 | 019303.jpg
 304 | 019304.jpg
 305 | 019305.jpg
 306 | 019306.jpg
 307 | 019307.jpg
 308 | 019308.jpg
 309 | 019309.jpg
 310 | 019310.jpg
 311 | 019311.jpg
 312 | 019312.jpg
 313 | 019313.jpg
 314 | 019314.jpg
 315 | 019315.jpg
 316 | 019316.jpg
 317 | 019317.jpg
 318 | 019318.jpg
 319 | 019319.jpg
 320 | 019320.jpg
 321 | 019321.jpg
 322 | 019322.jpg
 323 | 019323.jpg
 324 | 019324.jpg
 325 | 019325.jpg
 326 | 019326.jpg
 327 | 019327.jpg
 328 | 019328.jpg
 329 | 019329.jpg
 330 | 019330.jpg
 331 | 019331.jpg
 332 | 019332.jpg
 333 | 019333.jpg
 334 | 019334.jpg
 335 | 019335.jpg
 336 | 019336.jpg
 337 | 019337.jpg
 338 | 019338.jpg
 339 | 019339.jpg
 340 | 019340.jpg
 341 | 019341.jpg
 342 | 019342.jpg
 343 | 019343.jpg
 344 | 019344.jpg
 345 | 019345.jpg
 346 | 019346.jpg
 347 | 019347.jpg
 348 | 019348.jpg
 349 | 019349.jpg
 350 | 019350.jpg
 351 | 019351.jpg
 352 | 019352.jpg
 353 | 019353.jpg
 354 | 019354.jpg
 355 | 019355.jpg
 356 | 019356.jpg
 357 | 019357.jpg
 358 | 019358.jpg
 359 | 019359.jpg
 360 | 019360.jpg
 361 | 019361.jpg
 362 | 019362.jpg
 363 | 019363.jpg
 364 | 019364.jpg
 365 | 019365.jpg
 366 | 019366.jpg
 367 | 019367.jpg
 368 | 019368.jpg
 369 | 019369.jpg
 370 | 019370.jpg
 371 | 019371.jpg
 372 | 019372.jpg
 373 | 019373.jpg
 374 | 019374.jpg
 375 | 019375.jpg
 376 | 019376.jpg
 377 | 019377.jpg
 378 | 019378.jpg
 379 | 019379.jpg
 380 | 019380.jpg
 381 | 019381.jpg
 382 | 019382.jpg
 383 | 019383.jpg
 384 | 019384.jpg
 385 | 019385.jpg
 386 | 019386.jpg
 387 | 019387.jpg
 388 | 019388.jpg
 389 | 019389.jpg
 390 | 019390.jpg
 391 | 019391.jpg
 392 | 019392.jpg
 393 | 019393.jpg
 394 | 019394.jpg
 395 | 019395.jpg
 396 | 019396.jpg
 397 | 019397.jpg
 398 | 019398.jpg
 399 | 019399.jpg
 400 | 019400.jpg
 401 | 019401.jpg
 402 | 019402.jpg
 403 | 019403.jpg
 404 | 019404.jpg
 405 | 019405.jpg
 406 | 019406.jpg
 407 | 019407.jpg
 408 | 019408.jpg
 409 | 019409.jpg
 410 | 019410.jpg
 411 | 019411.jpg
 412 | 019412.jpg
 413 | 019413.jpg
 414 | 019414.jpg
 415 | 019415.jpg
 416 | 019416.jpg
 417 | 019417.jpg
 418 | 019418.jpg
 419 | 019419.jpg
 420 | 019420.jpg
 421 | 019421.jpg
 422 | 019422.jpg
 423 | 019423.jpg
 424 | 019424.jpg
 425 | 019425.jpg
 426 | 019426.jpg
 427 | 019427.jpg
 428 | 019428.jpg
 429 | 019429.jpg
 430 | 019430.jpg
 431 | 019431.jpg
 432 | 019432.jpg
 433 | 019433.jpg
 434 | 019434.jpg
 435 | 019435.jpg
 436 | 019436.jpg
 437 | 019437.jpg
 438 | 019438.jpg
 439 | 019439.jpg
 440 | 019440.jpg
 441 | 019441.jpg
 442 | 019442.jpg
 443 | 019443.jpg
 444 | 019444.jpg
 445 | 019445.jpg
 446 | 019446.jpg
 447 | 019447.jpg
 448 | 019448.jpg
 449 | 019449.jpg
 450 | 019450.jpg
 451 | 019451.jpg
 452 | 019452.jpg
 453 | 019453.jpg
 454 | 019454.jpg
 455 | 019455.jpg
 456 | 019456.jpg
 457 | 019457.jpg
 458 | 019458.jpg
 459 | 019459.jpg
 460 | 019460.jpg
 461 | 019461.jpg
 462 | 019462.jpg
 463 | 019463.jpg
 464 | 019464.jpg
 465 | 019465.jpg
 466 | 019466.jpg
 467 | 019467.jpg
 468 | 019468.jpg
 469 | 019469.jpg
 470 | 019470.jpg
 471 | 019471.jpg
 472 | 019472.jpg
 473 | 019473.jpg
 474 | 019474.jpg
 475 | 019475.jpg
 476 | 019476.jpg
 477 | 019477.jpg
 478 | 019478.jpg
 479 | 019479.jpg
 480 | 019480.jpg
 481 | 019481.jpg
 482 | 019482.jpg
 483 | 019483.jpg
 484 | 019484.jpg
 485 | 019485.jpg
 486 | 019486.jpg
 487 | 019487.jpg
 488 | 019488.jpg
 489 | 019489.jpg
 490 | 019490.jpg
 491 | 019491.jpg
 492 | 019492.jpg
 493 | 019493.jpg
 494 | 019494.jpg
 495 | 019495.jpg
 496 | 019496.jpg
 497 | 019497.jpg
 498 | 019498.jpg
 499 | 019499.jpg
 500 | 019500.jpg
 501 | 019501.jpg
 502 | 019502.jpg
 503 | 019503.jpg
 504 | 019504.jpg
 505 | 019505.jpg
 506 | 019506.jpg
 507 | 019507.jpg
 508 | 019508.jpg
 509 | 019509.jpg
 510 | 019510.jpg
 511 | 019511.jpg
 512 | 019512.jpg
 513 | 019513.jpg
 514 | 019514.jpg
 515 | 019515.jpg
 516 | 019516.jpg
 517 | 019517.jpg
 518 | 019518.jpg
 519 | 019519.jpg
 520 | 019520.jpg
 521 | 019521.jpg
 522 | 019522.jpg
 523 | 019523.jpg
 524 | 019524.jpg
 525 | 019525.jpg
 526 | 019526.jpg
 527 | 019527.jpg
 528 | 019528.jpg
 529 | 019529.jpg
 530 | 019530.jpg
 531 | 019531.jpg
 532 | 019532.jpg
 533 | 019533.jpg
 534 | 019534.jpg
 535 | 019535.jpg
 536 | 019536.jpg
 537 | 019537.jpg
 538 | 019538.jpg
 539 | 019539.jpg
 540 | 019540.jpg
 541 | 019541.jpg
 542 | 019542.jpg
 543 | 019543.jpg
 544 | 019544.jpg
 545 | 019545.jpg
 546 | 019546.jpg
 547 | 019547.jpg
 548 | 019548.jpg
 549 | 019549.jpg
 550 | 019550.jpg
 551 | 019551.jpg
 552 | 019552.jpg
 553 | 019553.jpg
 554 | 019554.jpg
 555 | 019555.jpg
 556 | 019556.jpg
 557 | 019557.jpg
 558 | 019558.jpg
 559 | 019559.jpg
 560 | 019560.jpg
 561 | 019561.jpg
 562 | 019562.jpg
 563 | 019563.jpg
 564 | 019564.jpg
 565 | 019565.jpg
 566 | 019566.jpg
 567 | 019567.jpg
 568 | 019568.jpg
 569 | 019569.jpg
 570 | 019570.jpg
 571 | 019571.jpg
 572 | 019572.jpg
 573 | 019573.jpg
 574 | 019574.jpg
 575 | 019575.jpg
 576 | 019576.jpg
 577 | 019577.jpg
 578 | 019578.jpg
 579 | 019579.jpg
 580 | 019580.jpg
 581 | 019581.jpg
 582 | 019582.jpg
 583 | 019583.jpg
 584 | 019584.jpg
 585 | 019585.jpg
 586 | 019586.jpg
 587 | 019587.jpg
 588 | 019588.jpg
 589 | 019589.jpg
 590 | 019590.jpg
 591 | 019591.jpg
 592 | 019592.jpg
 593 | 019593.jpg
 594 | 019594.jpg
 595 | 019595.jpg
 596 | 019596.jpg
 597 | 019597.jpg
 598 | 019598.jpg
 599 | 019599.jpg
 600 | 019600.jpg
 601 | 019601.jpg
 602 | 019602.jpg
 603 | 019603.jpg
 604 | 019604.jpg
 605 | 019605.jpg
 606 | 019606.jpg
 607 | 019607.jpg
 608 | 019608.jpg
 609 | 019609.jpg
 610 | 019610.jpg
 611 | 019611.jpg
 612 | 019612.jpg
 613 | 019613.jpg
 614 | 019614.jpg
 615 | 019615.jpg
 616 | 019616.jpg
 617 | 019617.jpg
 618 | 019618.jpg
 619 | 019619.jpg
 620 | 019620.jpg
 621 | 019621.jpg
 622 | 019622.jpg
 623 | 019623.jpg
 624 | 019624.jpg
 625 | 019625.jpg
 626 | 019626.jpg
 627 | 019627.jpg
 628 | 019628.jpg
 629 | 019629.jpg
 630 | 019630.jpg
 631 | 019631.jpg
 632 | 019632.jpg
 633 | 019633.jpg
 634 | 019634.jpg
 635 | 019635.jpg
 636 | 019636.jpg
 637 | 019637.jpg
 638 | 019638.jpg
 639 | 019639.jpg
 640 | 019640.jpg
 641 | 019641.jpg
 642 | 019642.jpg
 643 | 019643.jpg
 644 | 019644.jpg
 645 | 019645.jpg
 646 | 019646.jpg
 647 | 019647.jpg
 648 | 019648.jpg
 649 | 019649.jpg
 650 | 019650.jpg
 651 | 019651.jpg
 652 | 019652.jpg
 653 | 019653.jpg
 654 | 019654.jpg
 655 | 019655.jpg
 656 | 019656.jpg
 657 | 019657.jpg
 658 | 019658.jpg
 659 | 019659.jpg
 660 | 019660.jpg
 661 | 019661.jpg
 662 | 019662.jpg
 663 | 019663.jpg
 664 | 019664.jpg
 665 | 019665.jpg
 666 | 019666.jpg
 667 | 019667.jpg
 668 | 019668.jpg
 669 | 019669.jpg
 670 | 019670.jpg
 671 | 019671.jpg
 672 | 019672.jpg
 673 | 019673.jpg
 674 | 019674.jpg
 675 | 019675.jpg
 676 | 019676.jpg
 677 | 019677.jpg
 678 | 019678.jpg
 679 | 019679.jpg
 680 | 019680.jpg
 681 | 019681.jpg
 682 | 019682.jpg
 683 | 019683.jpg
 684 | 019684.jpg
 685 | 019685.jpg
 686 | 019686.jpg
 687 | 019687.jpg
 688 | 019688.jpg
 689 | 019689.jpg
 690 | 019690.jpg
 691 | 019691.jpg
 692 | 019692.jpg
 693 | 019693.jpg
 694 | 019694.jpg
 695 | 019695.jpg
 696 | 019696.jpg
 697 | 019697.jpg
 698 | 019698.jpg
 699 | 019699.jpg
 700 | 019700.jpg
 701 | 019701.jpg
 702 | 019702.jpg
 703 | 019703.jpg
 704 | 019704.jpg
 705 | 019705.jpg
 706 | 019706.jpg
 707 | 019707.jpg
 708 | 019708.jpg
 709 | 019709.jpg
 710 | 019710.jpg
 711 | 019711.jpg
 712 | 019712.jpg
 713 | 019713.jpg
 714 | 019714.jpg
 715 | 019715.jpg
 716 | 019716.jpg
 717 | 019717.jpg
 718 | 019718.jpg
 719 | 019719.jpg
 720 | 019720.jpg
 721 | 019721.jpg
 722 | 019722.jpg
 723 | 019723.jpg
 724 | 019724.jpg
 725 | 019725.jpg
 726 | 019726.jpg
 727 | 019727.jpg
 728 | 019728.jpg
 729 | 019729.jpg
 730 | 019730.jpg
 731 | 019731.jpg
 732 | 019732.jpg
 733 | 019733.jpg
 734 | 019734.jpg
 735 | 019735.jpg
 736 | 019736.jpg
 737 | 019737.jpg
 738 | 019738.jpg
 739 | 019739.jpg
 740 | 019740.jpg
 741 | 019741.jpg
 742 | 019742.jpg
 743 | 019743.jpg
 744 | 019744.jpg
 745 | 019745.jpg
 746 | 019746.jpg
 747 | 019747.jpg
 748 | 019748.jpg
 749 | 019749.jpg
 750 | 019750.jpg
 751 | 019751.jpg
 752 | 019752.jpg
 753 | 019753.jpg
 754 | 019754.jpg
 755 | 019755.jpg
 756 | 019756.jpg
 757 | 019757.jpg
 758 | 019758.jpg
 759 | 019759.jpg
 760 | 019760.jpg
 761 | 019761.jpg
 762 | 019762.jpg
 763 | 019763.jpg
 764 | 019764.jpg
 765 | 019765.jpg
 766 | 019766.jpg
 767 | 019767.jpg
 768 | 019768.jpg
 769 | 019769.jpg
 770 | 019770.jpg
 771 | 019771.jpg
 772 | 019772.jpg
 773 | 019773.jpg
 774 | 019774.jpg
 775 | 019775.jpg
 776 | 019776.jpg
 777 | 019777.jpg
 778 | 019778.jpg
 779 | 019779.jpg
 780 | 019780.jpg
 781 | 019781.jpg
 782 | 019782.jpg
 783 | 019783.jpg
 784 | 019784.jpg
 785 | 019785.jpg
 786 | 019786.jpg
 787 | 019787.jpg
 788 | 019788.jpg
 789 | 019789.jpg
 790 | 019790.jpg
 791 | 019791.jpg
 792 | 019792.jpg
 793 | 019793.jpg
 794 | 019794.jpg
 795 | 019795.jpg
 796 | 019796.jpg
 797 | 019797.jpg
 798 | 019798.jpg
 799 | 019799.jpg
 800 | 019800.jpg
 801 | 019801.jpg
 802 | 019802.jpg
 803 | 019803.jpg
 804 | 019804.jpg
 805 | 019805.jpg
 806 | 019806.jpg
 807 | 019807.jpg
 808 | 019808.jpg
 809 | 019809.jpg
 810 | 019810.jpg
 811 | 019811.jpg
 812 | 019812.jpg
 813 | 019813.jpg
 814 | 019814.jpg
 815 | 019815.jpg
 816 | 019816.jpg
 817 | 019817.jpg
 818 | 019818.jpg
 819 | 019819.jpg
 820 | 019820.jpg
 821 | 019821.jpg
 822 | 019822.jpg
 823 | 019823.jpg
 824 | 019824.jpg
 825 | 019825.jpg
 826 | 019826.jpg
 827 | 019827.jpg
 828 | 019828.jpg
 829 | 019829.jpg
 830 | 019830.jpg
 831 | 019831.jpg
 832 | 019832.jpg
 833 | 019833.jpg
 834 | 019834.jpg
 835 | 019835.jpg
 836 | 019836.jpg
 837 | 019837.jpg
 838 | 019838.jpg
 839 | 019839.jpg
 840 | 019840.jpg
 841 | 019841.jpg
 842 | 019842.jpg
 843 | 019843.jpg
 844 | 019844.jpg
 845 | 019845.jpg
 846 | 019846.jpg
 847 | 019847.jpg
 848 | 019848.jpg
 849 | 019849.jpg
 850 | 019850.jpg
 851 | 019851.jpg
 852 | 019852.jpg
 853 | 019853.jpg
 854 | 019854.jpg
 855 | 019855.jpg
 856 | 019856.jpg
 857 | 019857.jpg
 858 | 019858.jpg
 859 | 019859.jpg
 860 | 019860.jpg
 861 | 019861.jpg
 862 | 019862.jpg
 863 | 019863.jpg
 864 | 019864.jpg
 865 | 019865.jpg
 866 | 019866.jpg
 867 | 019867.jpg
 868 | 019868.jpg
 869 | 019869.jpg
 870 | 019870.jpg
 871 | 019871.jpg
 872 | 019872.jpg
 873 | 019873.jpg
 874 | 019874.jpg
 875 | 019875.jpg
 876 | 019876.jpg
 877 | 019877.jpg
 878 | 019878.jpg
 879 | 019879.jpg
 880 | 019880.jpg
 881 | 019881.jpg
 882 | 019882.jpg
 883 | 019883.jpg
 884 | 019884.jpg
 885 | 019885.jpg
 886 | 019886.jpg
 887 | 019887.jpg
 888 | 019888.jpg
 889 | 019889.jpg
 890 | 019890.jpg
 891 | 019891.jpg
 892 | 019892.jpg
 893 | 019893.jpg
 894 | 019894.jpg
 895 | 019895.jpg
 896 | 019896.jpg
 897 | 019897.jpg
 898 | 019898.jpg
 899 | 019899.jpg
 900 | 019900.jpg
 901 | 019901.jpg
 902 | 019902.jpg
 903 | 019903.jpg
 904 | 019904.jpg
 905 | 019905.jpg
 906 | 019906.jpg
 907 | 019907.jpg
 908 | 019908.jpg
 909 | 019909.jpg
 910 | 019910.jpg
 911 | 019911.jpg
 912 | 019912.jpg
 913 | 019913.jpg
 914 | 019914.jpg
 915 | 019915.jpg
 916 | 019916.jpg
 917 | 019917.jpg
 918 | 019918.jpg
 919 | 019919.jpg
 920 | 019920.jpg
 921 | 019921.jpg
 922 | 019922.jpg
 923 | 019923.jpg
 924 | 019924.jpg
 925 | 019925.jpg
 926 | 019926.jpg
 927 | 019927.jpg
 928 | 019928.jpg
 929 | 019929.jpg
 930 | 019930.jpg
 931 | 019931.jpg
 932 | 019932.jpg
 933 | 019933.jpg
 934 | 019934.jpg
 935 | 019935.jpg
 936 | 019936.jpg
 937 | 019937.jpg
 938 | 019938.jpg
 939 | 019939.jpg
 940 | 019940.jpg
 941 | 019941.jpg
 942 | 019942.jpg
 943 | 019943.jpg
 944 | 019944.jpg
 945 | 019945.jpg
 946 | 019946.jpg
 947 | 019947.jpg
 948 | 019948.jpg
 949 | 019949.jpg
 950 | 019950.jpg
 951 | 019951.jpg
 952 | 019952.jpg
 953 | 019953.jpg
 954 | 019954.jpg
 955 | 019955.jpg
 956 | 019956.jpg
 957 | 019957.jpg
 958 | 019958.jpg
 959 | 019959.jpg
 960 | 019960.jpg
 961 | 019961.jpg
 962 | 019962.jpg
 963 | 019963.jpg
 964 | 019964.jpg
 965 | 019965.jpg
 966 | 019966.jpg
 967 | 019967.jpg
 968 | 019968.jpg
 969 | 019969.jpg
 970 | 019970.jpg
 971 | 019971.jpg
 972 | 019972.jpg
 973 | 019973.jpg
 974 | 019974.jpg
 975 | 019975.jpg
 976 | 019976.jpg
 977 | 019977.jpg
 978 | 019978.jpg
 979 | 019979.jpg
 980 | 019980.jpg
 981 | 019981.jpg
 982 | 019982.jpg
 983 | 019983.jpg
 984 | 019984.jpg
 985 | 019985.jpg
 986 | 019986.jpg
 987 | 019987.jpg
 988 | 019988.jpg
 989 | 019989.jpg
 990 | 019990.jpg
 991 | 019991.jpg
 992 | 019992.jpg
 993 | 019993.jpg
 994 | 019994.jpg
 995 | 019995.jpg
 996 | 019996.jpg
 997 | 019997.jpg
 998 | 019998.jpg
 999 | 019999.jpg
1000 | 020000.jpg
1001 | 


--------------------------------------------------------------------------------
/o1.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/IntuitionMachines/OrigamiNet/9e82724cc2998d583a976b1ca781f7aeba73c38c/o1.png


--------------------------------------------------------------------------------
/requirements.txt:
--------------------------------------------------------------------------------
 1 | opencv_python_headless==4.1.2.30
 2 | natsort
 3 | numpy
 4 | tqdm
 5 | lmdb
 6 | scikit_image
 7 | matplotlib
 8 | torchsummary
 9 | nltk
10 | torchvision
11 | imageio
12 | six
13 | scipy
14 | fire
15 | Pillow
16 | scikit_learn
17 | horovod
18 | editdistance
19 | git+https://github.com/google/gin-config


--------------------------------------------------------------------------------
/test.py:
--------------------------------------------------------------------------------
  1 | import os
  2 | import time
  3 | import string
  4 | 
  5 | import torch
  6 | import torch.backends.cudnn as cudnn
  7 | import torch.utils.data
  8 | import torch.distributed as dist
  9 | import numpy as np
 10 | import editdistance
 11 | from nltk.translate.bleu_score import sentence_bleu
 12 | import horovod.torch as hvd
 13 | 
 14 | from utils import Averager
 15 | device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 16 | 
 17 | 
 18 | def metric_sum_hvd(val, name):
 19 |     tensor = torch.tensor(val)
 20 |     avg_tensor = hvd.allreduce(tensor, name=name, average=False)
 21 |     return avg_tensor.item()
 22 | 
 23 | def metric_sum_ddp(tensor, av=False):
 24 |     # rt = tensor.clone()
 25 |     rt = torch.tensor(tensor).cuda()
 26 |     dist.all_reduce(rt, op=dist.ReduceOp.SUM)
 27 |     if av:
 28 |         rt /= dist.get_world_size()
 29 |     return rt
 30 | 
 31 | def validation(model, criterion, evaluation_loader, converter, opt, parO):
 32 |     """ validation or evaluation """
 33 |     n_correct = 0
 34 |     norm_ED = 0
 35 |     tot_ED = 0
 36 |     length_of_gt = 0
 37 |     bleu = 0.0
 38 |     infer_time = 0
 39 |     valid_loss_avg = Averager()
 40 | 
 41 |     for i, (image_tensors, labels) in enumerate(evaluation_loader):
 42 |         batch_size = image_tensors.size(0)
 43 |         image = image_tensors.to(device)
 44 | 
 45 |         text_for_loss, length_for_loss = converter.encode(labels)
 46 | 
 47 |         start_time = time.time()
 48 | 
 49 |         preds = model(image, '')
 50 |         forward_time = time.time() - start_time
 51 | 
 52 |         # Calculate evaluation loss for CTC deocder.
 53 |         preds_size = torch.IntTensor([preds.size(1)] * batch_size)
 54 |         preds = preds.permute(1, 0, 2).log_softmax(2)  # to use CTCloss format
 55 | 
 56 |         # To avoid ctc_loss issue, disabled cudnn for the computation of the ctc_loss
 57 |         # https://github.com/jpuigcerver/PyLaia/issues/16
 58 |         torch.backends.cudnn.enabled = False
 59 |         cost = criterion(preds, text_for_loss, preds_size, length_for_loss).mean()
 60 |         torch.backends.cudnn.enabled = True
 61 | 
 62 |         _, preds_index = preds.max(2)
 63 |         preds_index = preds_index.transpose(1, 0).contiguous().view(-1)
 64 |         preds_str = converter.decode(preds_index.data, preds_size.data)
 65 | 
 66 |         infer_time += forward_time
 67 |         valid_loss_avg.add(cost)
 68 | 
 69 |         # calculate accuracy.
 70 |         for pred, gt in zip(preds_str, labels):
 71 |             tmped = editdistance.eval(pred, gt)
 72 |             if pred == gt:
 73 |                 n_correct += 1
 74 |             if len(gt) == 0:
 75 |                 norm_ED += 1
 76 |             else:
 77 |                 norm_ED += tmped / float(len(gt))
 78 | 
 79 |             tot_ED += tmped
 80 |             length_of_gt += len(gt)
 81 |             bleu += sentence_bleu( [list(gt)], list(pred) ) 
 82 | 
 83 |     if parO.HVD:
 84 |         n_correct  = metric_sum_hvd(n_correct , 'sum_n_correct')
 85 |         tot_ED  = metric_sum_hvd(tot_ED , 'sum_tot_ED')
 86 |         length_of_gt  = metric_sum_hvd(length_of_gt , 'sum_length_of_gt')
 87 |         norm_ED = metric_sum_hvd(norm_ED, 'sum_norm_ED')
 88 |         bleu = metric_sum_hvd(bleu, 'sum_bleu')
 89 | 
 90 |     elif parO.DDP:
 91 |         val_loss = metric_sum_ddp(valid_loss_avg.val(), av=True)
 92 |         n_correct  = metric_sum_ddp(n_correct)
 93 |         tot_ED  = metric_sum_ddp(tot_ED)
 94 |         length_of_gt  = metric_sum_ddp(length_of_gt)
 95 |         norm_ED = metric_sum_ddp(norm_ED)
 96 |         bleu = metric_sum_ddp(bleu)
 97 | 
 98 |     nelms = float(len(evaluation_loader.dataset))
 99 |     tot_ED = tot_ED / float(length_of_gt)
100 |     val_loss = valid_loss_avg.val() if not parO.DDP else val_loss
101 |     bleu /= nelms
102 |     norm_ED /= nelms
103 |     accuracy = n_correct / nelms * 100
104 | 
105 |     return val_loss, accuracy, norm_ED, tot_ED, bleu, preds_str, labels, infer_time


--------------------------------------------------------------------------------
/train.py:
--------------------------------------------------------------------------------
  1 | import os
  2 | import sys
  3 | import time
  4 | import random
  5 | import string
  6 | import argparse
  7 | from collections import namedtuple
  8 | import copy
  9 | 
 10 | import torch
 11 | import torch.backends.cudnn as cudnn
 12 | import torch.nn.init as init
 13 | import torch.optim as optim
 14 | import torch.utils.data
 15 | from torch import autograd
 16 | import torch.multiprocessing as mp
 17 | import torch.distributed as dist
 18 | from torch.utils.data import Dataset
 19 | from torch.nn.parallel import DistributedDataParallel as pDDP
 20 | 
 21 | from torchsummary import summary
 22 | from torchvision.utils import save_image
 23 | import horovod.torch as hvd
 24 | import gin
 25 | 
 26 | import numpy as np
 27 | from tqdm import tqdm, trange
 28 | from PIL import Image
 29 | 
 30 | import apex
 31 | from apex.parallel import DistributedDataParallel as aDDP
 32 | from apex.fp16_utils import *
 33 | from apex import amp
 34 | from apex.multi_tensor_apply import multi_tensor_applier
 35 | 
 36 | import wandb
 37 | import ds_load
 38 | 
 39 | from utils import CTCLabelConverter, Averager, ModelEma, Metric
 40 | from cnv_model import OrigamiNet, ginM
 41 | from test import validation
 42 | 
 43 | parOptions = namedtuple('parOptions', ['DP', 'DDP', 'HVD'])
 44 | parOptions.__new__.__defaults__ = (False,) * len(parOptions._fields)
 45 | 
 46 | pO = None
 47 | OnceExecWorker = None
 48 | 
 49 | device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 50 | 
 51 | def init_bn(model):
 52 |     if type(model) in [torch.nn.InstanceNorm2d, torch.nn.BatchNorm2d]:
 53 |         init.ones_(model.weight)
 54 |         init.zeros_(model.bias)
 55 | 
 56 |     elif type(model) in [torch.nn.Conv2d]:
 57 |         init.kaiming_uniform_(model.weight)
 58 |     
 59 | def WrkSeeder(_):
 60 |     return np.random.seed((torch.initial_seed()) % (2 ** 32))
 61 | 
 62 | @gin.configurable
 63 | def train(opt, AMP, WdB, train_data_path, train_data_list, test_data_path, test_data_list, experiment_name, 
 64 |             train_batch_size, val_batch_size, workers, lr, valInterval, num_iter, wdbprj, continue_model=''):
 65 | 
 66 |     HVD3P = pO.HVD or pO.DDP
 67 | 
 68 |     os.makedirs(f'./saved_models/{experiment_name}', exist_ok=True)
 69 | 
 70 |     if OnceExecWorker and WdB:
 71 |         wandb.init(project=wdbprj, name=experiment_name)
 72 |         wandb.config.update(opt)
 73 |     
 74 |     train_dataset = ds_load.myLoadDS(train_data_list, train_data_path)
 75 |     valid_dataset = ds_load.myLoadDS(test_data_list, test_data_path , ralph=train_dataset.ralph)
 76 | 
 77 |     if OnceExecWorker:
 78 |         print(pO)
 79 |         print('Alphabet :',len(train_dataset.alph),train_dataset.alph)
 80 |         for d in [train_dataset, valid_dataset]:
 81 |             print('Dataset Size :',len(d.fns))
 82 |             print('Max LbW : ',max(list(map(len,d.tlbls))) )
 83 |             print('#Chars : ',sum([len(x) for x in d.tlbls]))
 84 |             print('Sample label :',d.tlbls[-1])
 85 |             print("Dataset :", sorted(list(map(len,d.tlbls))) )
 86 |             print('-'*80)
 87 |     
 88 |     if opt.num_gpu > 1:
 89 |         workers = workers * ( 1 if HVD3P else opt.num_gpu )
 90 | 
 91 |     if HVD3P:
 92 |         train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset, num_replicas=opt.world_size, rank=opt.rank)
 93 |         valid_sampler = torch.utils.data.distributed.DistributedSampler(valid_dataset, num_replicas=opt.world_size, rank=opt.rank)
 94 | 
 95 |     train_loader  = torch.utils.data.DataLoader( train_dataset, batch_size=train_batch_size, shuffle=True if not HVD3P else False, 
 96 |                     pin_memory = True, num_workers = int(workers),
 97 |                     sampler = train_sampler if HVD3P else None,
 98 |                     worker_init_fn = WrkSeeder,
 99 |                     collate_fn = ds_load.SameTrCollate
100 |                 )
101 | 
102 |     valid_loader  = torch.utils.data.DataLoader( valid_dataset, batch_size=val_batch_size , pin_memory=True, 
103 |                     num_workers = int(workers), sampler=valid_sampler if HVD3P else None)
104 |     
105 |     model = OrigamiNet()
106 |     model.apply(init_bn)
107 |     model.train()
108 | 
109 |     if OnceExecWorker: import pprint;[print(k,model.lreszs[k]) for k in sorted(model.lreszs.keys())]
110 | 
111 |     biparams    = list(dict(filter(lambda kv: 'bias'     in kv[0], model.named_parameters())).values())
112 |     nonbiparams = list(dict(filter(lambda kv: 'bias' not in kv[0], model.named_parameters())).values())
113 | 
114 |     if not pO.DDP:
115 |         model = model.to(device)
116 |     else:
117 |         model.cuda(opt.rank)
118 |     
119 |     optimizer = optim.Adam(model.parameters(), lr=lr)
120 |     lr_scheduler = torch.optim.lr_scheduler.ExponentialLR(optimizer, gamma=10**(-1/90000))
121 | 
122 |     if OnceExecWorker and WdB:
123 |         wandb.watch(model, log="all")
124 | 
125 |     if pO.HVD:
126 |         hvd.broadcast_parameters(model.state_dict(), root_rank=0)
127 |         hvd.broadcast_optimizer_state(optimizer, root_rank=0)
128 |         optimizer = hvd.DistributedOptimizer(optimizer, named_parameters=model.named_parameters())
129 |         # optimizer = hvd.DistributedOptimizer(optimizer, named_parameters=model.named_parameters(), compression=hvd.Compression.fp16)
130 |     
131 |     if pO.DDP and opt.rank!=0:
132 |         random.seed()
133 |         np.random.seed()
134 | 
135 |     if AMP:
136 |         model, optimizer = amp.initialize(model, optimizer, opt_level = "O1")
137 |     if pO.DP:
138 |         model = torch.nn.DataParallel(model)
139 |     elif pO.DDP:
140 |         model = pDDP(model, device_ids=[opt.rank], output_device=opt.rank,find_unused_parameters=False)
141 | 
142 |     
143 |     
144 |     model_ema = ModelEma(model)
145 | 
146 |     if continue_model != '':
147 |         if OnceExecWorker: print(f'loading pretrained model from {continue_model}')
148 |         checkpoint = torch.load(continue_model, map_location=f'cuda:{opt.rank}' if HVD3P else None)
149 |         model.load_state_dict(checkpoint['model'], strict=True)
150 |         optimizer.load_state_dict(checkpoint['optimizer'])
151 |         model_ema._load_checkpoint(continue_model, f'cuda:{opt.rank}' if HVD3P else None)
152 | 
153 |     criterion = torch.nn.CTCLoss(reduction='none', zero_infinity=True).to(device)
154 |     converter = CTCLabelConverter(train_dataset.ralph.values())
155 |     
156 |     if OnceExecWorker:
157 |         with open(f'./saved_models/{experiment_name}/opt.txt', 'a') as opt_file:
158 |             opt_log = '------------ Options -------------\n'
159 |             args = vars(opt)
160 |             for k, v in args.items():
161 |                 opt_log += f'{str(k)}: {str(v)}\n'
162 |             opt_log += '---------------------------------------\n'
163 |             opt_log += gin.operative_config_str()
164 |             opt_file.write(opt_log)
165 |             if WdB:
166 |                 wandb.config.gin_str = gin.operative_config_str().splitlines()
167 | 
168 | 
169 |         print(optimizer)
170 |         print(opt_log)
171 | 
172 |     start_time = time.time()
173 |     best_accuracy = -1
174 |     best_norm_ED = 1e+6
175 |     best_CER = 1e+6
176 |     i = 0
177 |     gAcc = 1
178 |     epoch = 1
179 |     btReplay = False and AMP
180 |     max_batch_replays = 3
181 | 
182 |     if HVD3P: train_sampler.set_epoch(epoch)
183 |     titer = iter(train_loader)
184 | 
185 |     while(True):
186 |         start_time = time.time()
187 | 
188 |         model.zero_grad()
189 |         train_loss = Metric(pO,'train_loss')
190 | 
191 |         for j in trange(valInterval, leave=False, desc='Training'):
192 | 
193 |             try:
194 |                 image_tensors, labels = next(titer)
195 |             except StopIteration:
196 |                 epoch += 1
197 |                 if HVD3P: train_sampler.set_epoch(epoch)
198 |                 titer = iter(train_loader)
199 |                 image_tensors, labels = next(titer)
200 |                 
201 |             image = image_tensors.to(device)
202 |             text, length = converter.encode(labels)
203 |             batch_size = image.size(0)
204 | 
205 |             replay_batch = True
206 |             maxR = 3
207 |             while replay_batch and maxR>0:
208 |                 maxR -= 1
209 |                 
210 |                 preds = model(image,text).float()
211 |                 preds_size = torch.IntTensor([preds.size(1)] * batch_size).to(device)
212 |                 preds = preds.permute(1, 0, 2).log_softmax(2)
213 |                 
214 |                 if i==0 and OnceExecWorker:
215 |                     print('Model inp : ',image.dtype,image.size())
216 |                     print('CTC inp : ',preds.dtype,preds.size(),preds_size[0])
217 | 
218 |                 # To avoid ctc_loss issue, disabled cudnn for the computation of the ctc_loss
219 |                 torch.backends.cudnn.enabled = False
220 |                 cost = criterion(preds, text.to(device), preds_size, length.to(device)).mean() / gAcc
221 |                 torch.backends.cudnn.enabled = True
222 | 
223 |                 train_loss.update(cost)
224 | 
225 |                 optimizer.zero_grad()
226 |                 default_optimizer_step = optimizer.step  # added for batch replay
227 | 
228 |                 if not AMP:
229 |                     cost.backward()
230 |                     replay_batch = False
231 |                 else:
232 |                     with amp.scale_loss(cost, optimizer) as scaled_loss:
233 |                         scaled_loss.backward()
234 |                         if pO.HVD: optimizer.synchronize()
235 | 
236 |                     if optimizer.step is default_optimizer_step or not btReplay:
237 |                         replay_batch = False
238 |                     elif maxR>0:
239 |                         optimizer.step()
240 |                     
241 |                     
242 |             if btReplay: amp._amp_state.loss_scalers[0]._loss_scale = mx_sc
243 |             
244 |             if (i+1) % gAcc == 0:
245 | 
246 |                 if pO.HVD and AMP:
247 |                     with optimizer.skip_synchronize(): 
248 |                         optimizer.step()
249 |                 else:
250 |                     optimizer.step()
251 |                 
252 |                 model.zero_grad()
253 |                 model_ema.update(model, num_updates=i/2)
254 | 
255 |                 if (i+1) % (gAcc*2) == 0:
256 |                     lr_scheduler.step()
257 |             
258 |             i += 1
259 |             
260 |         # validation part
261 |         if True:
262 | 
263 |             elapsed_time = time.time() - start_time
264 |             start_time = time.time()
265 | 
266 |             model.eval()
267 |             with torch.no_grad():
268 | 
269 |                 
270 |                 valid_loss, current_accuracy, current_norm_ED, ted, bleu, preds, labels, infer_time = validation(
271 |                     model_ema.ema, criterion, valid_loader, converter, opt, pO)
272 |         
273 |             model.train()
274 |             v_time = time.time() - start_time
275 | 
276 |             if OnceExecWorker:                
277 |                 if current_norm_ED < best_norm_ED:
278 |                     best_norm_ED = current_norm_ED
279 |                     checkpoint = {
280 |                         'model': model.state_dict(),
281 |                         'state_dict_ema': model_ema.ema.state_dict(),
282 |                         'optimizer': optimizer.state_dict(),
283 |                     }
284 |                     torch.save(checkpoint, f'./saved_models/{experiment_name}/best_norm_ED.pth')
285 | 
286 |                 if ted < best_CER:
287 |                     best_CER = ted
288 |                 
289 |                 if current_accuracy > best_accuracy:
290 |                     best_accuracy = current_accuracy
291 | 
292 |                 out  = f'[{i}] Loss: {train_loss.avg:0.5f} time: ({elapsed_time:0.1f},{v_time:0.1f})'
293 |                 out += f' vloss: {valid_loss:0.3f}'
294 |                 out += f' CER: {ted:0.4f} NER: {current_norm_ED:0.4f} lr: {lr_scheduler.get_lr()[0]:0.5f}'
295 |                 out += f' bAcc: {best_accuracy:0.1f}, bNER: {best_norm_ED:0.4f}, bCER: {best_CER:0.4f}, B: {bleu*100:0.2f}'
296 |                 print(out)
297 | 
298 |                 with open(f'./saved_models/{experiment_name}/log_train.txt', 'a') as log: log.write(out + '\n')
299 | 
300 |                 if WdB:
301 |                     wandb.log({'lr': lr_scheduler.get_lr()[0], 'It':i, 'nED': current_norm_ED,  'B':bleu*100,
302 |                     'tloss':train_loss.avg, 'AnED': best_norm_ED, 'CER':ted, 'bestCER':best_CER, 'vloss':valid_loss})
303 | 
304 |         if i == num_iter:
305 |             print('end the training')
306 |             sys.exit()
307 | 
308 | def gInit(opt):
309 |     global pO, OnceExecWorker
310 |     gin.parse_config_file(opt.gin)
311 |     pO = parOptions(**{ginM('dist'):True})
312 | 
313 |     if pO.HVD:
314 |         hvd.init()
315 |         torch.cuda.set_device(hvd.local_rank())
316 | 
317 |     OnceExecWorker = (pO.HVD and hvd.rank() == 0) or (pO.DP)
318 |     cudnn.benchmark = True
319 | 
320 | 
321 | def rSeed(sd):
322 |     random.seed(sd)
323 |     np.random.seed(sd)
324 |     torch.manual_seed(sd)
325 |     torch.cuda.manual_seed(sd)
326 | 
327 | def launch_fn(rank, opt):
328 |     global OnceExecWorker
329 |     gInit(opt)
330 |     OnceExecWorker = OnceExecWorker or (pO.DDP and rank==0)
331 |     mp.set_start_method('fork', force=True)
332 | 
333 |     os.environ['MASTER_ADDR'] = 'localhost'
334 |     os.environ['MASTER_PORT'] = str(opt.port)
335 | 
336 |     dist.init_process_group("nccl", rank=rank, world_size=opt.num_gpu)
337 | 
338 |     #to ensure identical init parameters
339 |     rSeed(opt.manualSeed)
340 | 
341 |     torch.cuda.set_device(rank)
342 |     opt.world_size = opt.num_gpu
343 |     opt.rank       = rank
344 | 
345 |     train(opt)
346 | 
347 | if __name__ == '__main__':
348 | 
349 |     parser = argparse.ArgumentParser()
350 |     parser.add_argument('--gin', help='Gin config file')
351 | 
352 |     opt = parser.parse_args()
353 |     gInit(opt)
354 |     opt.manualSeed = ginM('manualSeed')
355 |     opt.port = ginM('port')
356 | 
357 |     if OnceExecWorker:
358 |         rSeed(opt.manualSeed)
359 | 
360 |     opt.num_gpu = torch.cuda.device_count()
361 |     
362 | 
363 |     if pO.HVD:
364 |         opt.world_size = hvd.size()
365 |         opt.rank       = hvd.rank()
366 |     
367 |     if not pO.DDP:
368 |         train(opt)
369 |     else:
370 |         mp.spawn(launch_fn, args=(opt,), nprocs=opt.num_gpu)


--------------------------------------------------------------------------------
/utils.py:
--------------------------------------------------------------------------------
  1 | import torch
  2 | import torch.distributed as dist
  3 | 
  4 | import horovod.torch as hvd
  5 | 
  6 | from copy import deepcopy
  7 | from collections import OrderedDict
  8 | device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
  9 | 
 10 | 
 11 | class CTCLabelConverter(object):
 12 |     """ Convert between text-label and text-index """
 13 | 
 14 |     def __init__(self, character):
 15 |         # character (str): set of the possible characters.
 16 |         dict_character = list(character)
 17 | 
 18 |         self.dict = {}
 19 |         for i, char in enumerate(dict_character):
 20 |             # NOTE: 0 is reserved for 'blank' token required by CTCLoss
 21 |             self.dict[char] = i + 1
 22 | 
 23 |         self.character = ['[blank]'] + dict_character  # dummy '[blank]' token for CTCLoss (index 0)
 24 | 
 25 |     def encode(self, text):
 26 |         """convert text-label into text-index.
 27 |         input:
 28 |             text: text labels of each image. [batch_size]
 29 | 
 30 |         output:
 31 |             text: concatenated text index for CTCLoss.
 32 |                     [sum(text_lengths)] = [text_index_0 + text_index_1 + ... + text_index_(n - 1)]
 33 |             length: length of each text. [batch_size]
 34 |         """
 35 |         length = [len(s) for s in text]
 36 |         text = ''.join(text)
 37 |         text = [self.dict[char] for char in text]
 38 | 
 39 |         return (torch.IntTensor(text).to(device), torch.IntTensor(length).to(device))
 40 | 
 41 |     def decode(self, text_index, length):
 42 |         """ convert text-index into text-label. """
 43 |         texts = []
 44 |         index = 0
 45 |         for l in length:
 46 |             t = text_index[index:index + l]
 47 | 
 48 |             char_list = []
 49 |             for i in range(l):
 50 |                 if t[i] != 0 and (not (i > 0 and t[i - 1] == t[i])) and t[i]<len(self.character):  # removing repeated characters and blank.
 51 |                     char_list.append(self.character[t[i]])
 52 |             text = ''.join(char_list)
 53 | 
 54 |             texts.append(text)
 55 |             index += l
 56 |         return texts
 57 | 
 58 | class Averager(object):
 59 |     """Compute average for torch.Tensor, used for loss average."""
 60 | 
 61 |     def __init__(self):
 62 |         self.reset()
 63 | 
 64 |     def add(self, v):
 65 |         count = v.data.numel()
 66 |         v = v.data.sum()
 67 |         self.n_count += count
 68 |         self.sum += v
 69 | 
 70 |     def reset(self):
 71 |         self.n_count = 0
 72 |         self.sum = 0
 73 | 
 74 |     def val(self):
 75 |         res = 0
 76 |         if self.n_count != 0:
 77 |             res = self.sum / float(self.n_count)
 78 |         return res
 79 | 
 80 | # Horovod: average metrics from distributed training.
 81 | class Metric(object):
 82 |     def __init__(self, parO, name=''):
 83 |         self.name = name
 84 |         self.sum = torch.tensor(0.).double()
 85 |         self.n = torch.tensor(0.)
 86 |         self.pO = parO
 87 | 
 88 |     def update(self, val):
 89 |         if self.pO.HVD:
 90 |             self.sum += hvd.allreduce(val.detach().cpu(), name=self.name).double()
 91 |         elif self.pO.DDP:
 92 |             rt = val.clone()
 93 |             dist.all_reduce(rt, op=dist.ReduceOp.SUM)
 94 |             rt /= dist.get_world_size()
 95 |             self.sum += rt.detach().cpu().double()
 96 |         elif self.pO.DP:
 97 |             self.sum += val.detach().double()
 98 | 
 99 |         self.n += 1
100 | 
101 |     @property
102 |     def avg(self):
103 |         return self.sum / self.n.double()
104 | 
105 | 
106 | #https://github.com/rwightman/pytorch-image-models/blob/ema-cleanup/utils.py
107 | class ModelEma:
108 |     """ Model Exponential Moving Average
109 |     Keep a moving average of everything in the model state_dict (parameters and buffers).
110 |     This is intended to allow functionality like
111 |     https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage
112 |     A smoothed version of the weights is necessary for some training schemes to perform well.
113 |     E.g. Google's hyper-params for training MNASNet, MobileNet-V3, EfficientNet, etc that use
114 |     RMSprop with a short 2.4-3 epoch decay period and slow LR decay rate of .96-.99 requires EMA
115 |     smoothing of weights to match results. Pay attention to the decay constant you are using
116 |     relative to your update count per epoch.
117 |     To keep EMA from using GPU resources, set device='cpu'. This will save a bit of memory but
118 |     disable validation of the EMA weights. Validation will have to be done manually in a separate
119 |     process, or after the training stops converging.
120 |     This class is sensitive where it is initialized in the sequence of model init,
121 |     GPU assignment and distributed training wrappers.
122 |     I've tested with the sequence in my own train.py for torch.DataParallel, apex.DDP, and single-GPU.
123 |     """
124 |     def __init__(self, model, decay=0.9999, device='', resume=''):
125 |         # make a copy of the model for accumulating moving average of weights
126 |         self.ema = deepcopy(model)
127 |         self.ema.eval()
128 |         self.decay = decay
129 |         self.device = device  # perform ema on different device from model if set
130 |         if device:
131 |             self.ema.to(device=device)
132 |         self.ema_has_module = hasattr(self.ema, 'module')
133 |         if resume:
134 |             self._load_checkpoint(resume)
135 |         for p in self.ema.parameters():
136 |             p.requires_grad_(False)
137 | 
138 |     def _load_checkpoint(self, checkpoint_path, mapl=None):
139 |         checkpoint = torch.load(checkpoint_path,map_location=mapl)
140 |         assert isinstance(checkpoint, dict)
141 |         if 'state_dict_ema' in checkpoint:
142 |             new_state_dict = OrderedDict()
143 |             for k, v in checkpoint['state_dict_ema'].items():
144 |                 # ema model may have been wrapped by DataParallel, and need module prefix
145 |                 if self.ema_has_module:
146 |                     name = 'module.' + k if not k.startswith('module') else k
147 |                 else:
148 |                     name = k
149 |                 new_state_dict[name] = v
150 |             self.ema.load_state_dict(new_state_dict)
151 |             print("=> Loaded state_dict_ema")
152 |         else:
153 |             print("=> Failed to find state_dict_ema, starting from loaded model weights")
154 | 
155 |     def update(self, model, num_updates=-1):
156 |         # correct a mismatch in state dict keys
157 |         needs_module = hasattr(model, 'module') and not self.ema_has_module
158 |         if num_updates >= 0:
159 |             _cdecay = min(self.decay, (1 + num_updates) / (10 + num_updates))
160 |         else:
161 |             _cdecay = self.decay
162 | 
163 |         with torch.no_grad():
164 |             msd = model.state_dict()
165 |             for k, ema_v in self.ema.state_dict().items():
166 |                 if needs_module:
167 |                     k = 'module.' + k
168 |                 model_v = msd[k].detach()
169 |                 if self.device:
170 |                     model_v = model_v.to(device=self.device)
171 |                 ema_v.copy_(ema_v * _cdecay + (1. - _cdecay) * model_v)


--------------------------------------------------------------------------------