├── LICENSE
├── README.md
├── cls_models
    ├── __init__.py
    └── model.py
├── cpp_inference
    ├── .vscode
    │   └── settings.json
    ├── CMakeLists.txt
    ├── classification
    │   ├── CMakeLists.txt
    │   ├── include
    │   │   └── ImgCls.hpp
    │   ├── src
    │   │   └── ImgCls.cpp
    │   └── test_ImgCls.cpp
    ├── compile.sh
    └── traced_model
    │   └── trace_model.py
├── dataset
    ├── __init__.py
    ├── cls_dataset.py
    └── transform.py
├── ensamble
    ├── kaggle_vote.py
    └── samples
    │   ├── method_1.csv
    │   ├── method_2.csv
    │   └── method_3.csv
├── run.sh
├── sample_files
    └── imgs
    │   ├── cat
    │       ├── 0.jpg
    │       ├── 1.jpg
    │       ├── 2.jpg
    │       ├── 3.jpg
    │       ├── 4.jpg
    │       └── 5.jpg
    │   ├── dog
    │       ├── 0.jpg
    │       ├── 1.jpg
    │       ├── 2.jpg
    │       ├── 3.jpg
    │       ├── 4.jpg
    │       └── 5.jpg
    │   └── listfile.txt
├── tools
    ├── evaluation.py
    ├── predict.py
    └── train_val.py
├── trt_inference
    ├── CMakeLists.txt
    ├── convert_onnx.py
    ├── include
    │   └── TrtCls.hpp
    ├── labels.txt
    └── src
    │   └── TrtCls.cpp
├── utils
    ├── __init__.py
    ├── label_smoothing_pytorch.py
    ├── loss_kd.py
    ├── lr_scheduler.py
    ├── opts.py
    ├── util.py
    └── warmup_lr.py
└── visualization
    ├── Feature_Visualization.ipynb
    ├── Feature_Visualization.py
    ├── f1_conv1.png
    └── test.png


/LICENSE:
--------------------------------------------------------------------------------
 1 | MIT License
 2 | 
 3 | Copyright (c) 2020 xiangzhe_lu
 4 | 
 5 | Permission is hereby granted, free of charge, to any person obtaining a copy
 6 | of this software and associated documentation files (the "Software"), to deal
 7 | in the Software without restriction, including without limitation the rights
 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 9 | copies of the Software, and to permit persons to whom the Software is
10 | furnished to do so, subject to the following conditions:
11 | 
12 | The above copyright notice and this permission notice shall be included in all
13 | copies or substantial portions of the Software.
14 | 
15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 | SOFTWARE.
22 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | 
 2 | ## 简介
 3 | 
 4 | 基于torchision实现的pytorch图像分类功能。
 5 | 
 6 | 
 7 | ## 近期更新
 8 | 
 9 | * 2022.11.05更新
10 |     - 新添加tensorrt c++的推理方案
11 | 
12 | * 2022.10.29更新，进行代码重构，基本的功能基本一致。
13 |     - 支持pytorch ddp的训练
14 |     - 支持c++ libtorch的模型推理
15 |     - 支持script脚本一键运行
16 |     - 添加日志模块
17 | 
18 | 习惯之前版本的请看v1版本的代码：[V1版本](https://github.com/lxztju/pytorch_classification/tree/v1)。
19 | 
20 | 
21 | 主要功能：
22 | 
23 | 利用pytorch实现图像分类，基于torchision可以扩展使用densenet，resnext，mobilenet，efficientnet，swin transformer等图像分类网络
24 | 
25 | 如果有用欢迎star
26 | 
27 | ## 实现功能
28 | * 基础功能利用pytorch实现图像分类
29 | * 包含带有warmup的cosine学习率调整
30 | * warmup的step学习率优调整
31 | * 多模型融合预测，加权与投票融合
32 | * 利用flask + redis实现模型云端api部署（tag v1）
33 | * c++ libtorch的模型部署
34 | * 使用tta测试时增强进行预测（tag v1）
35 | * 添加label smooth的pytorch实现（标签平滑）（tag v1）
36 | * 添加使用cnn提取特征，并使用SVM，RF，MLP，KNN等分类器进行分类（tag v1）。
37 | * 可视化特征层
38 | 
39 | ## 运行环境
40 | * python3.7
41 | * pytorch 1.8.1
42 | * torchvision 0.9.1
43 | * opencv(libtorch cpp推理使用， 版本3.4.6)（可选）
44 | * libtorch cpp推理使用（可选）
45 | 
46 | 
47 | 
48 | ## 快速开始
49 | 
50 | ### 数据集形式
51 |  数据集的组织形式，参考[sample_files/imgs/listfile.txt](https://github.com/lxztju/pytorch_classification/blob/master/sample_files/imgs/listfile.txt)
52 | 
53 | 
54 | ### 训练 测试
55 | 
56 | 修改`run.sh`中的参数，直接运行run.sh即可运行
57 | 
58 | 
59 | 主要修改的参数：
60 | 
61 | ```
62 | OUTPUT_PATH 模型保存和log文件的路径
63 | 
64 | TRAIN_LIST 训练数据集的list文件
65 | VAL_LIST  测试集合的list文件
66 | model_name 默认是resnet50
67 | lr 学习率
68 | epochs 训练总的epoch
69 | batch-size  batch的大小
70 | j dataloader的num_workers的大小
71 | num_classes 类别数
72 | ```
73 | 
74 | 
75 | ### libtorch inference
76 | 
77 | 
78 | 代码存储在`cpp_inference`文件夹中。
79 | 
80 | 1. 利用[cpp_inference/traced_model/trace_model.py](https://github.com/lxztju/pytorch_classification/blob/master/cpp_inference/traced_model/trace_model.py)将训练好的模型导出。
81 | 2. 编译所需的opencv和libtorch代码到`cpp_inference/third_party_library`
82 | 
83 | 3. 编译
84 | ```
85 | sh compile.sh
86 | ```
87 | 
88 | 4. 可执行文件测试
89 | ```
90 | ./bin/imgCls imgpath
91 | ```
92 | 
93 | 
94 | 


--------------------------------------------------------------------------------
/cls_models/__init__.py:
--------------------------------------------------------------------------------
1 | from .model import ClsModel


--------------------------------------------------------------------------------
/cls_models/model.py:
--------------------------------------------------------------------------------
 1 | from logging import raiseExceptions
 2 | import torch
 3 | import torch.nn as nn
 4 | import torchvision
 5 | 
 6 | ModelWeights = {
 7 |     'mobilenet_v2':'MobileNet_V2_Weights.IMAGENET1K_V1',
 8 |     'resnet18':'ResNet18_Weights.IMAGENET1K_V1',
 9 |     'resnet50':'ResNet50_Weights.IMAGENET1K_V1',
10 |     'resnet101' : 'ResNet101_Weights.IMAGENET1K_V1',
11 |     'swin_s': 'Swin_S_Weights.IMAGENET1K_V1',
12 |     'swin_b': 'Swin_B_Weights.IMAGENET1K_V1',
13 |     'vit_b_16': 'ViT_B_16_Weights.IMAGENET1K_V1',
14 |     'vit_b_32' : 'ViT_B_32_Weights.IMAGENET1K_V1',
15 |     'vit_l_16': 'ViT_L_16_Weights.IMAGENET1K_V1',
16 |     'vit_l_32': 'ViT_L_32_Weights.IMAGENET1K_V1'
17 | }
18 | 
19 | 
20 | class ClsModel(nn.Module):
21 |     def __init__(self, model_name, num_classes, is_pretrained=False):
22 |         super(ClsModel, self).__init__()
23 |         self.model_name = model_name
24 |         self.num_class = num_classes
25 |         self.is_pretrained = is_pretrained
26 | 
27 |         if self.model_name not in ModelWeights:
28 |             raise ValueError('Please confirm the name of model')
29 | 
30 |         if self.is_pretrained:
31 |             self.base_model = getattr(torchvision.models, self.model_name)(weights=ModelWeights[self.model_name])
32 |         else:
33 |             self.base_model = getattr(torchvision.models, self.model_name)()
34 | 
35 |         if hasattr(self.base_model, 'classifier'):
36 |             self.base_model.last_layer_name = 'classifier'
37 |             feature_dim = self.base_model.classifier[1].in_features
38 |             self.base_model.classifier = nn.Linear(feature_dim, self.num_class)
39 |         elif hasattr(self.base_model, 'fc'):
40 |             self.base_model.last_layer_name = 'fc'
41 |             feature_dim = getattr(self.base_model, self.base_model.last_layer_name).in_features
42 |             self.base_model.fc = nn.Linear(feature_dim, self.num_class)
43 |         elif hasattr(self.base_model, 'head'):
44 |             self.base_model.last_layer_name = 'head'
45 |             feature_dim = getattr(self.base_model, self.base_model.last_layer_name).in_features
46 |             self.base_model.head = nn.Linear(feature_dim, self.num_class)
47 |         elif hasattr(self.base_model, 'heads'):
48 |             self.base_model.last_layer_name = 'heads'
49 |             feature_dim = getattr(self.base_model, self.base_model.last_layer_name).in_features
50 |             self.base_model.heads = nn.Linear(feature_dim, self.num_class)
51 |         else:
52 |             raise ValueError('Please confirm the name of last')
53 | 
54 | #         self.new_fc = nn.Linear(feature_dim, self.num_class)
55 | 
56 | 
57 |     def forward(self, x):
58 |         x = self.base_model(x)
59 | #         x = self.new_fc(x)
60 |         return x
61 | 
62 | 
63 | if __name__ == '__main__':
64 |     model_name = 'resnet50'
65 |     num_classes = 2
66 |     is_pretrained = False
67 | 
68 |     clsmodel = ClsModel(model_name, num_classes, 0, is_pretrained)
69 |     print(clsmodel)
70 | 


--------------------------------------------------------------------------------
/cpp_inference/.vscode/settings.json:
--------------------------------------------------------------------------------
 1 | {
 2 |     "files.associations": {
 3 |         "array": "cpp",
 4 |         "atomic": "cpp",
 5 |         "bit": "cpp",
 6 |         "*.tcc": "cpp",
 7 |         "cctype": "cpp",
 8 |         "clocale": "cpp",
 9 |         "cmath": "cpp",
10 |         "compare": "cpp",
11 |         "concepts": "cpp",
12 |         "cstdarg": "cpp",
13 |         "cstddef": "cpp",
14 |         "cstdint": "cpp",
15 |         "cstdio": "cpp",
16 |         "cstdlib": "cpp",
17 |         "cwchar": "cpp",
18 |         "cwctype": "cpp",
19 |         "deque": "cpp",
20 |         "string": "cpp",
21 |         "unordered_map": "cpp",
22 |         "vector": "cpp",
23 |         "exception": "cpp",
24 |         "algorithm": "cpp",
25 |         "functional": "cpp",
26 |         "iterator": "cpp",
27 |         "memory": "cpp",
28 |         "memory_resource": "cpp",
29 |         "numeric": "cpp",
30 |         "random": "cpp",
31 |         "string_view": "cpp",
32 |         "system_error": "cpp",
33 |         "tuple": "cpp",
34 |         "type_traits": "cpp",
35 |         "utility": "cpp",
36 |         "initializer_list": "cpp",
37 |         "iosfwd": "cpp",
38 |         "limits": "cpp",
39 |         "new": "cpp",
40 |         "numbers": "cpp",
41 |         "ostream": "cpp",
42 |         "stdexcept": "cpp",
43 |         "streambuf": "cpp",
44 |         "typeinfo": "cpp",
45 |         "iostream": "cpp",
46 |         "istream": "cpp"
47 |     }
48 | }


--------------------------------------------------------------------------------
/cpp_inference/CMakeLists.txt:
--------------------------------------------------------------------------------
 1 | cmake_minimum_required(VERSION 3.5)
 2 | project(cpp_deployment)
 3 | 
 4 | add_definitions(-D _GLIBCXX_USE_CXX11_ABI=1)
 5 | 
 6 | set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/lib)
 7 | set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/lib)
 8 | set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/bin)
 9 | 
10 | 
11 | set(CMAKE_BUILD_TYPE Release)
12 | 
13 | set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
14 | 
15 | set(CLASS_3DPARTY ${CMAKE_CURRENT_SOURCE_DIR}/third_party_library)
16 | 
17 | 
18 | # 两个必要的依赖包
19 | set(OpenCV_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party_library/opencv-3.4.16/share/OpenCV)
20 | set(Torch_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party_library/libtorch/share/cmake/Torch)
21 | find_package(OpenCV REQUIRED)
22 | find_package(Torch REQUIRED)
23 | 
24 | message(STATUS "OpenCV_LIBS: ${OpenCV_LIBS}")
25 | message(STATUS "Torch_LIBS: ${TORCH_LIBRARIES}")
26 | 
27 | 
28 | 
29 | ## build 
30 | add_subdirectory(classification)


--------------------------------------------------------------------------------
/cpp_inference/classification/CMakeLists.txt:
--------------------------------------------------------------------------------
 1 | cmake_minimum_required(VERSION 3.5)
 2 | project(classification)
 3 | 
 4 | set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
 5 | 
 6 | 
 7 | 
 8 | include_directories(
 9 |     ${TORCH_INCLUDE_DIRS}
10 |     ${OpenCV_INCLUDE_DIRS}
11 |     ${CMAKE_CURRENT_SOURCE_DIR}/include
12 | )
13 | 
14 | message(STATUS "Torch_INCLUDE_DIRS: ${TORCH_INCLUDE_DIRS}")
15 | # link_directories(
16 | #     ${OpenCV_LIBRARIES_DIRS}
17 | #     # ${OpenCV_DIR}/lib
18 | #     ${Torch_LIBRARIES_DIRS}
19 | # )
20 | 
21 | 
22 | # link_directories(
23 | #     ${OpenCV_LIBRARIES_DIRS}
24 | # )
25 | 
26 | # file (GLOB CPP_SRC src/*.cpp)
27 | 
28 | add_library(classification SHARED src/ImgCls.cpp)
29 | target_link_libraries(
30 |     classification
31 |     ${TORCH_LIBRARIES}
32 |     ${OpenCV_LIBS}
33 | )
34 | 
35 | 
36 | add_executable(imgcls test_ImgCls.cpp)
37 | target_link_libraries(
38 |     imgcls
39 |     classification
40 |     ${TORCH_LIBRARIES}
41 |     ${OpenCV_LIBS}
42 | )


--------------------------------------------------------------------------------
/cpp_inference/classification/include/ImgCls.hpp:
--------------------------------------------------------------------------------
 1 | #ifndef IMAGE_CLS_H
 2 | #define IMAGE_CLS_H
 3 | 
 4 | #include <string>
 5 | #include <vector>
 6 | #include <torch/torch.h>
 7 | #include <torch/script.h>
 8 | 
 9 | #include <opencv2/opencv.hpp>
10 | 
11 | class ImgCls{
12 | public:
13 |     ImgCls();
14 |     ~ImgCls();
15 | public:
16 |     int Init(std::string model_path, std::string labelmap_path);
17 |     int run_img(std::string img_path);
18 |     int run(std::vector<cv::Mat>& input_Mats, std::vector<cv::Mat>& output_Mats);
19 | 
20 | 
21 | private:
22 | 
23 |     torch::jit::script::Module cls_model;
24 |     std::vector<std::string> labels;
25 |     std::vector<float> thresh;
26 |     bool debug_ = true;
27 |     int cls_width = 224;
28 |     int cls_height = 224;
29 |     cv::Scalar mean = (0.485, 0.456, 0.406);
30 |     cv::Scalar std = (0.229, 0.224, 0.225);
31 | };
32 | #endif


--------------------------------------------------------------------------------
/cpp_inference/classification/src/ImgCls.cpp:
--------------------------------------------------------------------------------
 1 | #include "ImgCls.hpp"
 2 | 
 3 | #include <iostream>
 4 | #include<string>
 5 | #include <torch/torch.h>
 6 | #include <torch/script.h>
 7 | 
 8 | #include <opencv2/opencv.hpp>
 9 | 
10 | 
11 | // 构造和析构函数的定义
12 | ImgCls::ImgCls(){
13 | 
14 | }
15 | 
16 | ImgCls::~ImgCls(){
17 | 
18 | }
19 | 
20 | 
21 | // 模型初始化的定义
22 | int ImgCls::Init(std::string model_path, std::string labelmap_path){
23 | 
24 |     // 模型初始化
25 |     cls_model = torch::jit::load(model_path);
26 |     if (torch::cuda::is_available()){
27 |         torch::Device device0(torch::kCUDA);
28 |         cls_model.to(device0);
29 |     }
30 |     else{
31 |         torch::Device device0(torch::kCPU);
32 |         cls_model.to(device0);
33 |     }
34 |     cls_model.eval();
35 |     std::cout<<"cls model init done"<<std::endl;
36 | 
37 |     //idx2label的初始化。 txt文件，文件内容为： idx \t label_name
38 | 
39 |     return 0;
40 | }
41 | 
42 | 
43 | int ImgCls::run_img(std::string img_path){
44 |     cv::Mat image_ori = cv::imread(img_path);
45 |     cv::Mat image;
46 |     cv::cvtColor(image_ori, image, cv::COLOR_BGR2RGB);
47 |     cv::resize(image, image, cv::Size(224, 224));
48 |     image.convertTo(image, CV_32F, 1.0 / 255);
49 |     
50 |     //scaled, then subtract mean and div std.
51 |     cv::subtract(image, mean, image);
52 |     cv::divide(image, std, image);
53 |             
54 |     torch::Tensor img_tensor = torch::from_blob(image.data, {1, image.rows, image.cols,3}, torch::kFloat);
55 | 
56 |     img_tensor = img_tensor.permute({0, 3, 1, 2}).contiguous();
57 |     
58 |     std::vector<torch::jit::IValue> inputs;
59 |     if(torch::cuda::is_available()){  
60 |         torch::Device device0(torch::kCUDA);
61 |         inputs.push_back(img_tensor.to(device0));
62 |     }
63 |     else{
64 |         torch::Device device0(torch::kCPU);
65 |         inputs.push_back(img_tensor.to(device0));
66 |     }
67 |     torch::Tensor output = cls_model.forward(inputs).toTensor();
68 | 
69 |     auto output_scores = torch::softmax(output, 1);
70 |     if (debug_){
71 |         std::cout<<"torch::Tensor output: "<<output.type() <<" " << output<<std::endl;
72 |     }
73 |     
74 |     
75 |     std::tuple<torch::Tensor, torch::Tensor> max_result = torch::max(output_scores, 1);
76 | 
77 |     at::Tensor max_scores = std::get<0>(max_result);
78 |     at::Tensor max_indexs = std::get<1>(max_result);
79 |     if (debug_){
80 |         std::cout<<"max_score: "<<max_scores<<std::endl;
81 |         std::cout<<"max_index: "<<max_indexs<<std::endl;
82 |     }
83 |     // tensor转为标准类型
84 |     float scores = max_scores.item<float>();
85 |     int indexs = max_indexs.item<int>();
86 |     return indexs;
87 | }
88 | 
89 | 
90 | int ImgCls::run(std::vector<cv::Mat>& input_Mats, std::vector<cv::Mat>& output_Mats){
91 |     return 0;
92 | }


--------------------------------------------------------------------------------
/cpp_inference/classification/test_ImgCls.cpp:
--------------------------------------------------------------------------------
 1 | #include "ImgCls.hpp"
 2 | 
 3 | #include <iostream>
 4 | #include <torch/torch.h>
 5 | #include <torch/script.h>
 6 | 
 7 | #include <opencv2/opencv.hpp>
 8 | 
 9 | 
10 | 
11 | 
12 | 
13 | int main(int argc, const char** argv){
14 |     std::string img_path = argv[1];
15 |     
16 |     std::string model_path = "/home/lxz/codes/pytorch_classification/cpp_inference/traced_model/traced_model_res50.pt";
17 |     std::string labelmap_path = "";
18 |     ImgCls imgcls;
19 |     imgcls.Init(model_path, labelmap_path);
20 |     auto res = imgcls.run_img(img_path);
21 |     std::cout<<"res: "<<res<<std::endl;
22 | }


--------------------------------------------------------------------------------
/cpp_inference/compile.sh:
--------------------------------------------------------------------------------
1 | if [ ! -d "./build/" ];then
2 |     mkdir build 
3 | fi
4 | cd build &&cmake -DCMAKE_CXX_STANDARD=14 .. && make -j2


--------------------------------------------------------------------------------
/cpp_inference/traced_model/trace_model.py:
--------------------------------------------------------------------------------
 1 | from operator import is_
 2 | import torch
 3 | import sys 
 4 | sys.path.append('/home/lxz/pytorch_classification')
 5 | print(sys.path)
 6 | from cls_models import ClsModel
 7 | 
 8 | 
 9 | model = ClsModel('resnet50', num_classes=2, dropout=0, is_pretrained=False)
10 | 
11 | sd = torch.load('./trained_model.pth', map_location='cpu')
12 | model.load_state_dict(sd)
13 | model.eval()
14 | example = torch.rand(1, 3, 224, 224)
15 | 
16 | 
17 | traced_script_module = torch.jit.trace(model, example)
18 | # traced_script_module.eval()
19 | traced_script_module.save('./traced_model_res50.pt')
20 | output = traced_script_module(torch.ones(1, 3, 224, 224))
21 | 
22 | 
23 | traced_model_load = torch.jit.load('./traced_model_res50.pt')
24 | output1 = traced_model_load(torch.ones(1, 3, 224, 224))
25 | 
26 | print(output)
27 | print(output1)


--------------------------------------------------------------------------------
/dataset/__init__.py:
--------------------------------------------------------------------------------
1 | from .cls_dataset import ClsDataset
2 | from .transform import train_transform, val_transform


--------------------------------------------------------------------------------
/dataset/cls_dataset.py:
--------------------------------------------------------------------------------
 1 | import os
 2 | from PIL import Image
 3 | 
 4 | import torch
 5 | from torch.utils.data import Dataset
 6 | import numpy as np
 7 | 
 8 | class ImageInfo:
 9 |     def __init__(self, row):
10 |         self._data = row
11 | 
12 |     @property
13 |     def path(self):
14 |         return self._data[0]
15 | 
16 |     @property
17 |     def label(self):
18 |         return int(self._data[1])
19 | 
20 | 
21 | 
22 | class  ClsDataset(Dataset):
23 |     def __init__(self, list_file, transform=None):
24 |         self.list_file = list_file
25 |         self.transform = transform
26 |         
27 |         self._parser_input_data()
28 |                  
29 |     def _parser_input_data(self):
30 |         assert os.path.exists(self.list_file)
31 |         
32 |         lines = [x.strip().split('\t') for x in open(self.list_file, encoding='utf-8')]
33 |         
34 |         self.imgs_list = [ImageInfo(line) for line in lines]
35 |         
36 |         
37 | 
38 |     def __getitem__(self, index):
39 |         img_info = self.imgs_list[index]
40 |     
41 |         img = Image.open(img_info.path).convert('RGB')
42 |         if self.transform != None:
43 |             img =self.transform(img)
44 |         return img, torch.as_tensor(img_info.label, dtype=torch.long), img_info.path
45 |  
46 |     def __len__(self):
47 |         return len(self.imgs_list)
48 | 
49 | if __name__ == '__main__':
50 |     clsdataset = ClsDataset(list_file='list.txt')
51 |     for k, v in clsdataset:
52 |         print(k, v)
53 | 


--------------------------------------------------------------------------------
/dataset/transform.py:
--------------------------------------------------------------------------------
 1 | import random
 2 | 
 3 | 
 4 | from PIL import Image, ImageFilter
 5 | from torchvision import transforms
 6 | 
 7 | class Resize(object):
 8 |     def __init__(self, size, interpolation=Image.BILINEAR):
 9 |         self.size = size
10 |         self.interpolation = interpolation
11 | 
12 |     def __call__(self, img):
13 |         # padding
14 |         ratio = self.size[0] / self.size[1]
15 |         w, h = img.size
16 |         if w / h < ratio:
17 |             t = int(h * ratio)
18 |             w_padding = (t - w) // 2
19 |             img = img.crop((-w_padding, 0, w+w_padding, h))
20 |         else:
21 |             t = int(w / ratio)
22 |             h_padding = (t - h) // 2
23 |             img = img.crop((0, -h_padding, w, h+h_padding))
24 | 
25 |         img = img.resize(self.size, self.interpolation)
26 | 
27 |         return img
28 | 
29 | class RandomRotate(object):
30 |     def __init__(self, degree, p=0.5):
31 |         self.degree = degree
32 |         self.p = p
33 | 
34 |     def __call__(self, img):
35 |         if random.random() < self.p:
36 |             rotate_degree = random.uniform(-1*self.degree, self.degree)
37 |             img = img.rotate(rotate_degree, Image.BILINEAR)
38 |         return img
39 | 
40 | class RandomGaussianBlur(object):
41 |     def __init__(self, p=0.5):
42 |         self.p = p
43 |     def __call__(self, img):
44 |         if random.random() < self.p:
45 |             img = img.filter(ImageFilter.GaussianBlur(
46 |                 radius=random.random()))
47 |         return img
48 | 
49 | 
50 | 
51 | mean, std = [0.485, 0.456, 0.406], [0.229, 0.224, 0.225]
52 | def train_transform(mean=mean, std=std, size=0):
53 |     return transforms.Compose([
54 |         Resize((int(size * (256 / 224)), int(size * (256 / 224)))),
55 |         transforms.RandomCrop(size),
56 |         transforms.RandomVerticalFlip(),
57 |         transforms.RandomHorizontalFlip(),
58 |         RandomRotate(15, 0.3),
59 |         # RandomGaussianBlur(),
60 |         transforms.ToTensor(),
61 |         transforms.Normalize(mean=mean, std=std),
62 |     ])
63 | 
64 | def val_transform(mean=mean, std=std, size=0):
65 |     return transforms.Compose([
66 |         Resize((int(size * (256 / 224)), int(size * (256 / 224)))),
67 |         transforms.CenterCrop(size),
68 |         transforms.ToTensor(),
69 |         transforms.Normalize(mean=mean, std=std),
70 |     ])
71 | 


--------------------------------------------------------------------------------
/ensamble/kaggle_vote.py:
--------------------------------------------------------------------------------
 1 | # 源码来自于 https://github.com/MLWave/Kaggle-Ensemble-Guide
 2 | from collections import defaultdict, Counter
 3 | from glob import glob
 4 | import sys
 5 | import re
 6 | 
 7 | # sys.argv 就是运行程序时，在命令行输入的命令
 8 | # 运行程序时输入：python ./kaggle_vote.py "./samples/method*.csv" "./samples/vote.csv"
 9 | # 因此 sys.argv返回['./kaggle_vote.py','./samples/method*.csv', './samples/vote.csv']
10 | # 如果运行程序时采用：python ./kaggle_vote.py "./samples/method*.csv" "./samples/vote_weighted.csv" "weighted"
11 | # sys.argv返回['./kaggle_vote.py','./samples/method*.csv', './samples/vote_weighted.csv', 'weighted']
12 | 
13 | glob_files = sys.argv[1]
14 | loc_outfile = sys.argv[2]
15 | weights_strategy = "uniform"
16 | if len(sys.argv) == 4:
17 |   weights_strategy = sys.argv[3]
18 | 
19 | def kaggle_bag(glob_files, loc_outfile, method="average", weights="uniform"):
20 |   pattern = re.compile(r"(.)*_[w|W](\d*)_[.]*")
21 |   if method == "average":
22 |     scores = defaultdict(list)
23 |   with open(loc_outfile,"w") as outfile:
24 |     #weight_list may be usefull using a different method
25 |     weight_list = [1]*len(glob(glob_files))
26 |     for i, glob_file in enumerate( glob(glob_files) ):
27 |       print("parsing: {}".format(glob_file))
28 |       if weights == "weighted":
29 |          weight = pattern.match(glob_file)
30 |          if weight and weight.group(2):
31 |             print("Using weight: {}".format(weight.group(2)))
32 |             weight_list[i] = weight_list[i]*int(weight.group(2))
33 |          else:
34 |             print("Using weight: 1")
35 |       # sort glob_file by first column, ignoring the first line
36 |       lines = open(glob_file).readlines()
37 |       lines = [lines[0]] + sorted(lines[1:])
38 |       for e, line in enumerate( lines ):
39 |         if i == 0 and e == 0:
40 |           outfile.write(line)
41 |         if e > 0:
42 |           row = line.strip().split(",")
43 |           for l in range(1,weight_list[i]+1):
44 |             scores[(e,row[0])].append(row[1])
45 |     for j,k in sorted(scores):
46 |       outfile.write("%s,%s\n"%(k,Counter(scores[(j,k)]).most_common(1)[0][0]))
47 |     print("wrote to {}".format(loc_outfile))
48 | 
49 | kaggle_bag(glob_files, loc_outfile, weights=weights_strategy)
50 | 


--------------------------------------------------------------------------------
/ensamble/samples/method_2.csv:
--------------------------------------------------------------------------------
   1 | 04U72oFk9wo_03023.jpg,97
   2 | 04U72oFk9wo_03665.jpg,97
   3 | DJQ-_zmc-f0_01790.jpg,97
   4 | XjAmNbuGRl0_02032.jpg,97
   5 | SQqgkLfUC0A_03000.jpg,97
   6 | creOEIrJKII_00084.jpg,8
   7 | XjAmNbuGRl0_01425.jpg,97
   8 | 04U72oFk9wo_05471.jpg,97
   9 | tuFjRSDwOjw_03864.jpg,50
  10 | YQD_j5XOk_8_04184.jpg,97
  11 | khjc3IZSrEo_05661.jpg,50
  12 | 38qcxshnM54_00592.jpg,97
  13 | 6UG1-5QHcZo_01688.jpg,97
  14 | 2m0Ybw9yWX8_06699.jpg,8
  15 | hRw8lrLyzoE_00366.jpg,97
  16 | iDlLCKnOB2M_13400.jpg,8
  17 | J3ioFXV_fgE_02540.jpg,8
  18 | 9c2Atsc4JnM_01430.jpg,97
  19 | XjAmNbuGRl0_04150.jpg,97
  20 | dBS9kMQ7tcc_00383.jpg,50
  21 | XjAmNbuGRl0_01038.jpg,97
  22 | 04U72oFk9wo_03491.jpg,97
  23 | 04U72oFk9wo_00068.jpg,97
  24 | 04U72oFk9wo_05450.jpg,97
  25 | I5aNCGWztB4_00363.jpg,8
  26 | 4eg-kK6FeRQ_01037.jpg,97
  27 | lb5ycPZ0RUg_05515.jpg,97
  28 | 2m0Ybw9yWX8_07305.jpg,97
  29 | hMnu4nY8Ip8_01699.jpg,50
  30 | DrpNgJU3lwI_05168.jpg,97
  31 | _0E8-cthRM8_03382.jpg,97
  32 | FRA5wEek8Co_07799.jpg,97
  33 | 5Dle76vhHNM_00517.jpg,50
  34 | iDlLCKnOB2M_10649.jpg,8
  35 | hMnu4nY8Ip8_02706.jpg,97
  36 | XjAmNbuGRl0_01021.jpg,97
  37 | J3ioFXV_fgE_03779.jpg,97
  38 | hMnu4nY8Ip8_02851.jpg,97
  39 | 9c2Atsc4JnM_06100.jpg,97
  40 | ptVhsZ9KAk8_07578.jpg,8
  41 | DCzzgLvofA4_01357.jpg,97
  42 | 04U72oFk9wo_00340.jpg,97
  43 | XjAmNbuGRl0_03194.jpg,97
  44 | RVVvq3D50S8_04603.jpg,97
  45 | kC_pGKYXHYk_01769.jpg,0
  46 | tG5Wg_WT7T0_01922.jpg,0
  47 | LjfTvZ-cmzs_05278.jpg,0
  48 | VE90_U6vwAk_06751.jpg,0
  49 | ox6YzyVQww0_06934.jpg,0
  50 | JmcOzqZljK8_03320.jpg,0
  51 | VLk0NVNgW3Y_02022.jpg,0
  52 | -HgvlKE_Z4o_12262.jpg,0
  53 | mNsJopxZ7Po_01194.jpg,0
  54 | YJ9RPPmv_HQ_09016.jpg,0
  55 | GlxPOLCc3BY_01670.jpg,0
  56 | zDnRdinNrmE_02427.jpg,0
  57 | U6n4w_F0s5o_01170.jpg,0
  58 | 683nPQV_HT8_04796.jpg,0
  59 | I6o21B7Y6VQ_01325.jpg,0
  60 | PRmLt3ITUtA_00125.jpg,0
  61 | opsphwapsyg_04753.jpg,0
  62 | o1BXeQfPn1Y_03690.jpg,0
  63 | twWaRT4Rz8E_01235.jpg,0
  64 | X5NsLbPL_VI_05541.jpg,0
  65 | UuHbgKcUo7w_02044.jpg,0
  66 | jL8vZJ_om_M_03711.jpg,0
  67 | lgPOqmTuvKg_11711.jpg,0
  68 | NLJ5V6DxpQI_01709.jpg,0
  69 | 2KvrCuegt-g_02996.jpg,0
  70 | jlWJg0G60UE_07600.jpg,0
  71 | wwkX2HBtg-8_02860.jpg,0
  72 | ji9iEXiAyJ0_06955.jpg,0
  73 | IF-f4UNLL9o_13423.jpg,0
  74 | ank7LyVZ4M8_00375.jpg,0
  75 | uug0bhnLgHs_01611.jpg,0
  76 | Pqb0E2cxBbw_05457.jpg,0
  77 | L1kuD0u5RTo_00045.jpg,0
  78 | QkGyNAgEazo_03558.jpg,0
  79 | 3tSfgIa_hG0_03089.jpg,0
  80 | 6S7dnV3LenA_00725.jpg,0
  81 | LBnS47B0kK4_00788.jpg,0
  82 | v38ODWKoULY_09723.jpg,0
  83 | VMFYjVIsC9k_02557.jpg,0
  84 | _nGpX5ck078_03913.jpg,0
  85 | zDnRdinNrmE_08573.jpg,0
  86 | _zKtPVAf1Lc_05133.jpg,0
  87 | SXxzLMXbCQ0_03774.jpg,0
  88 | PQSO1Bd5avM_09628.jpg,0
  89 | HWcpjw2-jOo_03617.jpg,0
  90 | DzCp-F6a5Hg_02987.jpg,0
  91 | UDP7dzDzXUE_01035.jpg,0
  92 | tc07_SRevLs_03716.jpg,0
  93 | drmMzAe13XU_02111.jpg,0
  94 | EQbN3ftEkDI_03793.jpg,0
  95 | QUmBqvMv89E_04540.jpg,0
  96 | mNsJopxZ7Po_01133.jpg,0
  97 | AbqJEe6lLU0_00694.jpg,0
  98 | PE7KHQWD8K4_01975.jpg,0
  99 | Vl_hBjLk-A4_02698.jpg,0
 100 | H5Xb1iDpSkw_02881.jpg,0
 101 | Pn-JQRcREww_04190.jpg,0
 102 | nhcSsX1BNgA_06827.jpg,0
 103 | a2bJHDaswlk_08202.jpg,0
 104 | bs8Tw-lALHk_00780.jpg,0
 105 | vrQac74JRKc_13841.jpg,0
 106 | lvQy2r4dLMU_03398.jpg,0
 107 | aksSoUHp_Tg_00100.jpg,0
 108 | I8ckIq2-j0k_02810.jpg,0
 109 | ZY6Ac6YlYrA_00944.jpg,0
 110 | a2bJHDaswlk_00429.jpg,0
 111 | nrzfRt_h1ZA_06961.jpg,0
 112 | olieCr8lwxs_04600.jpg,0
 113 | H1WgQPvOJkI_01470.jpg,0
 114 | mp0OhGkC58E_00501.jpg,0
 115 | duZpdSKteu8_07833.jpg,0
 116 | Ba-IHHwT_eg_08326.jpg,0
 117 | ZY6Ac6YlYrA_08373.jpg,0
 118 | IeqQuZRYn58_04600.jpg,0
 119 | cFiHQZsZnlI_01802.jpg,0
 120 | wQat_Uv8ND4_03105.jpg,0
 121 | xPxzrGLvyxc_09201.jpg,0
 122 | 0GhX-stqtCU_06489.jpg,0
 123 | qK1oqP7oreA_03592.jpg,0
 124 | MP6F7nuOA9o_00831.jpg,0
 125 | qHQWCXwoLqA_01890.jpg,0
 126 | 5qDEXBm5O-c_00000.jpg,0
 127 | 5WbSYa9SSfc_02762.jpg,0
 128 | L1kuD0u5RTo_03249.jpg,0
 129 | L1kuD0u5RTo_02885.jpg,0
 130 | WOvFCGjGp1A_09957.jpg,0
 131 | P0N1G0Pj7ww_08617.jpg,0
 132 | ohzLXITapZA_10552.jpg,0
 133 | AgN67ee_Xlw_05130.jpg,0
 134 | NcN7Y7wS9Qk_00234.jpg,0
 135 | 3N-oR7JZbxQ_05182.jpg,0
 136 | h9qsdH9a6fA_06624.jpg,0
 137 | K3dXoygBIj0_01263.jpg,0
 138 | cFiHQZsZnlI_00387.jpg,0
 139 | cxee_3-KfDU_00668.jpg,0
 140 | 9yqTJ37IVjc_02456.jpg,0
 141 | vOfLkR1gamo_07900.jpg,0
 142 | tWaycPEUEjA_01293.jpg,0
 143 | NcyuAt-18cs_10297.jpg,0
 144 | N4d__h1KhZw_11719.jpg,0
 145 | VE90_U6vwAk_04426.jpg,0
 146 | aiIpolk4FGc_05875.jpg,0
 147 | js3Ph7SrPEE_01112.jpg,0
 148 | L-ke2ESp39g_01429.jpg,0
 149 | CG2GQ1mQTl8_04030.jpg,0
 150 | _vC4VBtMw08_01648.jpg,0
 151 | dGo6vicsm00_07116.jpg,0
 152 | XBElqDYUHVE_13731.jpg,0
 153 | zDnRdinNrmE_03360.jpg,0
 154 | fIeGAZg1oVA_07429.jpg,0
 155 | RBEJXMKfvhM_02191.jpg,0
 156 | AbvB3yxCimo_03386.jpg,0
 157 | GAlDrfZdIEE_01506.jpg,0
 158 | EUr4fitcCUM_00316.jpg,0
 159 | Nd4mWOpYwG0_09620.jpg,0
 160 | 6QQiveo7dG8_05355.jpg,0
 161 | v3bp5QQ7tAs_01197.jpg,0
 162 | SXxzLMXbCQ0_03712.jpg,0
 163 | tXS0RdlqhTo_04990.jpg,0
 164 | kgh4hgA92v0_07093.jpg,0
 165 | vEojnW7J3AQ_03950.jpg,0
 166 | fsXco4AQ2Lw_01622.jpg,0
 167 | FwRLT4Y2Ly0_01265.jpg,0
 168 | YJ9RPPmv_HQ_05077.jpg,0
 169 | FsUC567Oq0E_06790.jpg,0
 170 | Wxc904f21JI_03562.jpg,95
 171 | cqh1oeOWgP8_07672.jpg,0
 172 | qmIQSe4dSZk_00237.jpg,0
 173 | MTpF6mRks-o_01153.jpg,0
 174 | AbvB3yxCimo_03287.jpg,0
 175 | OL_bDVVxmwA_17239.jpg,0
 176 | iqcKPtuY68g_08999.jpg,0
 177 | XgkfqIrDeS4_06935.jpg,0
 178 | 3CGM2IFOfjA_05306.jpg,0
 179 | zMvNsAhhw2M_00370.jpg,0
 180 | i7kOOpYlLUk_02787.jpg,0
 181 | BBEEBUdSFqg_00188.jpg,0
 182 | QWXlKD-XGCQ_06677.jpg,0
 183 | 2HOIITIk5Ks_01050.jpg,0
 184 | i01d2joH9kQ_03063.jpg,0
 185 | PQSO1Bd5avM_01552.jpg,0
 186 | _hM7h-Iyp3Q_03642.jpg,0
 187 | kPU9fRvKQw0_04784.jpg,0
 188 | W9GseiFweqc_09848.jpg,0
 189 | Z6KavxJTyaM_03159.jpg,0
 190 | 0GhX-stqtCU_03629.jpg,0
 191 | CF9Q5gm0-wU_04386.jpg,0
 192 | 1Pw_ed5Nzjw_01718.jpg,0
 193 | mnFa0QAe0Og_04372.jpg,0
 194 | pvMQOW44vMU_00621.jpg,144
 195 | KJGOzpyGiJA_00554.jpg,0
 196 | KJGOzpyGiJA_02964.jpg,0
 197 | ORSV2Yk5EsQ_01947.jpg,0
 198 | AinaDOWuLEE_04825.jpg,0
 199 | M0W4xaj7szI_03660.jpg,0
 200 | 4AL-bBBkrsk_04899.jpg,0
 201 | 0-jEJ3ZVOyU_00203.jpg,0
 202 | Ld2N4OcNRAQ_02370.jpg,0
 203 | RBEJXMKfvhM_01623.jpg,0
 204 | wdasrVE5NOc_07629.jpg,0
 205 | tSNvmr41RSQ_03445.jpg,0
 206 | mTqyfjC6LjI_00083.jpg,0
 207 | 4siAdlKp9VE_05311.jpg,0
 208 | o1BXeQfPn1Y_04112.jpg,0
 209 | YFvqdld7ciM_01672.jpg,0
 210 | jRNDKz_aUMw_02277.jpg,0
 211 | 872cCNYHCsM_03925.jpg,0
 212 | YKXvjIq4wCU_03929.jpg,0
 213 | FQ8dPVTiXfQ_08303.jpg,0
 214 | wdasrVE5NOc_06252.jpg,0
 215 | 2cRvAaFHhV4_00776.jpg,0
 216 | XgkfqIrDeS4_07714.jpg,0
 217 | ox6YzyVQww0_08728.jpg,0
 218 | bFOEBL9xFw4_02249.jpg,0
 219 | Hy9FRK9fdaw_11410.jpg,0
 220 | h9PQeseW_LI_00340.jpg,0
 221 | UVoGwYghiU8_07725.jpg,0
 222 | fZoqLSHdwa8_00846.jpg,0
 223 | QGcjKx-JS64_02997.jpg,95
 224 | uGvRAA6O3pw_03962.jpg,0
 225 | lo8TtxZELYY_00397.jpg,0
 226 | ncfTX1cnb5k_09952.jpg,0
 227 | lW-LunvDQ_Y_06500.jpg,0
 228 | ZSzE0xJoh0Q_02270.jpg,0
 229 | LETOZQ5jPw0_01546.jpg,0
 230 | AbvB3yxCimo_01670.jpg,0
 231 | vs-3iBOtI5E_09013.jpg,0
 232 | qbtGbVK9jm0_03656.jpg,0
 233 | ZY6Ac6YlYrA_04088.jpg,0
 234 | gy46GY3erPM_06468.jpg,0
 235 | kOFVuvAYV7Q_00419.jpg,0
 236 | Y2ikHLeYPN0_04560.jpg,0
 237 | _O9CexLDFw4_00429.jpg,0
 238 | 1p_s-6esVJE_05066.jpg,0
 239 | 6KB3FYKE-g4_07624.jpg,0
 240 | 5vaFpvjaY8Q_02196.jpg,0
 241 | DX7qwr6hPxA_01427.jpg,0
 242 | cX5sUNF2pI4_03757.jpg,0
 243 | A_gcDgmF-8o_01702.jpg,57
 244 | N4d__h1KhZw_14692.jpg,0
 245 | xAatwq9U3wk_03840.jpg,0
 246 | zQjYHfRwgaA_02884.jpg,0
 247 | L8oSZ6NK6ss_05905.jpg,0
 248 | uqdzTEcVGEE_02849.jpg,0
 249 | NcyuAt-18cs_00742.jpg,0
 250 | UXxc4Ywkkdw_02783.jpg,0
 251 | P0N1G0Pj7ww_08833.jpg,0
 252 | zuD70UXuzhg_14015.jpg,0
 253 | jiPCEsoeCYs_10710.jpg,0
 254 | 1Z6s7F5j3Fs_04159.jpg,0
 255 | MaKkN8dQBu8_07935.jpg,0
 256 | BKUwqwXZXWc_08360.jpg,0
 257 | 4ReIPDy4H7w_04338.jpg,0
 258 | U653DzkSi3g_00029.jpg,0
 259 | jYplNjyvbJY_00116.jpg,0
 260 | B8fylxLu0Ic_02938.jpg,0
 261 | 9x7-GYIxMjI_03267.jpg,0
 262 | C7VZ0wXMEWs_08348.jpg,0
 263 | aCZ4PH8JTR4_02032.jpg,0
 264 | tjy7wieW0zo_00481.jpg,0
 265 | ib3yzPmYcmk_07841.jpg,0
 266 | vZzl0RbIQPM_01178.jpg,0
 267 | 1jZGc95fx9I_09547.jpg,0
 268 | aksSoUHp_Tg_01224.jpg,0
 269 | yuOkyxmxtf8_10384.jpg,0
 270 | QtABwMy2Vkk_07094.jpg,0
 271 | uNDrJ-arx4o_00413.jpg,0
 272 | r-6_FpiJzXM_04833.jpg,0
 273 | BL-9GGXn1SU_05448.jpg,0
 274 | SjMbvYVt0e4_04030.jpg,0
 275 | C7VZ0wXMEWs_07275.jpg,0
 276 | qXrgFO1c9yI_01957.jpg,0
 277 | H0jWhFNMpsI_01028.jpg,0
 278 | 6QQiveo7dG8_13554.jpg,0
 279 | sxLRPnBjoFI_00918.jpg,0
 280 | 6eiWBZlwGTI_02253.jpg,0
 281 | 0_7I3FTLrVQ_13467.jpg,0
 282 | gDFbBYWwUo0_02635.jpg,0
 283 | Jfa6FUzKDRk_04275.jpg,0
 284 | E9Vn6Hpg2W4_03634.jpg,0
 285 | U1KreouGo3M_07954.jpg,0
 286 | LETOZQ5jPw0_05123.jpg,0
 287 | GH0fuMUJbeY_02003.jpg,0
 288 | J1CYgyY9NWo_05561.jpg,0
 289 | reP2ugO6kjE_12661.jpg,0
 290 | jL8vZJ_om_M_05820.jpg,0
 291 | CVgiNJHlkFI_01230.jpg,0
 292 | reP2ugO6kjE_01737.jpg,0
 293 | vs-3iBOtI5E_08451.jpg,0
 294 | n9UX24MUWy0_08492.jpg,0
 295 | cFiHQZsZnlI_09421.jpg,0
 296 | eB-Joq8qxps_10344.jpg,0
 297 | 5Ep8acEixUU_00949.jpg,0
 298 | YFVibKse-5o_00604.jpg,0
 299 | 0BbmXmXLROA_00818.jpg,0
 300 | vs-3iBOtI5E_07960.jpg,0
 301 | 0GhX-stqtCU_09753.jpg,0
 302 | I8ckIq2-j0k_04622.jpg,0
 303 | h7NMRZ2rq30_04865.jpg,0
 304 | 9H3gCli6Sdo_02169.jpg,0
 305 | YyvfspBx_IY_07012.jpg,0
 306 | ZY6Ac6YlYrA_06830.jpg,0
 307 | lu9RPI5KpQo_05715.jpg,0
 308 | OiJx6tAXMik_05558.jpg,0
 309 | 6RRleFx9PzM_11295.jpg,0
 310 | 7-XMY9_6ZqA_00752.jpg,0
 311 | 21JpcjlyWiU_01124.jpg,0
 312 | oUgutUMibI8_01461.jpg,0
 313 | _-nqq2M2Y20_01359.jpg,0
 314 | qs_b9XksWrQ_02641.jpg,0
 315 | sLVRzsrZs74_02192.jpg,0
 316 | aCZ4PH8JTR4_10815.jpg,0
 317 | BPCaReAsUoQ_01278.jpg,0
 318 | hIiIzRVgXIE_01156.jpg,0
 319 | iVTFvRR8gsE_07247.jpg,0
 320 | GBZRXWezBHM_04285.jpg,0
 321 | qXrgFO1c9yI_02028.jpg,0
 322 | hQ3eeJyGr70_05334.jpg,0
 323 | 7ypQUG6SQQA_03823.jpg,0
 324 | 0RlyhYcoj2Y_04419.jpg,0
 325 | 683nPQV_HT8_00087.jpg,0
 326 | TGY3iP4X34c_06584.jpg,0
 327 | fQ7uyWBXN4s_00160.jpg,0
 328 | DbCY7UZBN7A_02095.jpg,0
 329 | 96A2lGSOOZM_00849.jpg,0
 330 | JbcRzZLC9-U_03977.jpg,0
 331 | 6SMyxbqxFZ8_04792.jpg,0
 332 | 1zAVpK5bRck_04213.jpg,0
 333 | w_GBd5YJtFU_08133.jpg,0
 334 | DNMfoG8o94U_01694.jpg,0
 335 | VO8FSYHjSKg_03317.jpg,0
 336 | z-2w7gWCXRY_15242.jpg,26
 337 | 42Tg2lSzo-M_00192.jpg,0
 338 | ulXBtXprT90_02841.jpg,0
 339 | RBEJXMKfvhM_01429.jpg,0
 340 | gdon6vuOC80_03853.jpg,0
 341 | K0WIuvbwS4o_03526.jpg,0
 342 | 8H8P8fAW8ns_21000.jpg,0
 343 | V48bYhdRjxY_05471.jpg,0
 344 | 5_wl7CD5B7Q_00644.jpg,0
 345 | puE2idkg7CQ_03784.jpg,0
 346 | JmcOzqZljK8_01797.jpg,0
 347 | K3dXoygBIj0_08381.jpg,0
 348 | h9PQeseW_LI_02386.jpg,0
 349 | HWcpjw2-jOo_06860.jpg,0
 350 | 99SRJCS2Egg_05679.jpg,0
 351 | PCVdAItDjxo_01997.jpg,0
 352 | k-2paW223h4_07407.jpg,0
 353 | 3YsP2SX4BDU_10331.jpg,0
 354 | 1Z6s7F5j3Fs_01444.jpg,0
 355 | xAatwq9U3wk_05537.jpg,0
 356 | DfDZt3dSuwY_02271.jpg,0
 357 | RcjW2zw4bYI_04827.jpg,0
 358 | wRxAl6EO2WY_07028.jpg,0
 359 | GEO1vdNUGWo_02095.jpg,0
 360 | sJLapskS7KU_06345.jpg,0
 361 | qKAT9HQCyeo_01472.jpg,0
 362 | jSh6VJYRV-g_01543.jpg,0
 363 | TOBkb-AOR8Q_07531.jpg,0
 364 | l5ju8C2Kku4_01195.jpg,0
 365 | l5ju8C2Kku4_07396.jpg,0
 366 | Vthw1qBE_C8_08640.jpg,0
 367 | RJIWbnJ-A10_04324.jpg,0
 368 | QO6dSqZykXs_04345.jpg,0
 369 | 3ORC9otWGMk_00102.jpg,0
 370 | 0VVP5TmxDZY_04396.jpg,0
 371 | h6sKA9Nmys4_06675.jpg,0
 372 | EORjg5P2DJs_03471.jpg,2
 373 | AGSx9mS14dw_02913.jpg,0
 374 | 5_wl7CD5B7Q_05476.jpg,0
 375 | JYTCwNspJsg_05436.jpg,0
 376 | 5fgyTbHvPDI_01454.jpg,0
 377 | 6KB3FYKE-g4_11222.jpg,0
 378 | WAUpelESAXo_05224.jpg,0
 379 | nOK3ytqRBpU_02496.jpg,0
 380 | SjMbvYVt0e4_08803.jpg,0
 381 | 4vLjW2SrFM0_00000.jpg,197
 382 | PxJRhNzDwYY_04718.jpg,0
 383 | 10u4S1a6leU_03437.jpg,0
 384 | 2RM963apVBw_01392.jpg,0
 385 | CV5DjqDmVes_05576.jpg,0
 386 | HttbrZul3x0_03777.jpg,0
 387 | qTTxnKk4pOo_00050.jpg,0
 388 | fIeGAZg1oVA_05360.jpg,0
 389 | 0GhX-stqtCU_01820.jpg,0
 390 | Pqb0E2cxBbw_05752.jpg,0
 391 | CWJo-_6S_sc_06283.jpg,0
 392 | RMdkMsov9W0_03862.jpg,0
 393 | sjI5Ig59tMI_03126.jpg,0
 394 | cp0pJDL5zhE_05996.jpg,0
 395 | biyaeWSbt5E_00570.jpg,0
 396 | COROzGYA84U_01152.jpg,2
 397 | gZuDMKXWU_E_05232.jpg,0
 398 | TO-Suwy1KAE_05052.jpg,0
 399 | tq2CwGgz73Q_04633.jpg,0
 400 | _v3Yb8Bzxp0_08615.jpg,0
 401 | dW-zA4iqTK4_04338.jpg,0
 402 | VgdvthNxlFI_02940.jpg,0
 403 | uF_yaCZI920_08190.jpg,0
 404 | isvBLbvn3n0_01180.jpg,0
 405 | W9GseiFweqc_07444.jpg,0
 406 | AinaDOWuLEE_02017.jpg,0
 407 | cFiHQZsZnlI_11025.jpg,0
 408 | KBuIYWRO-sM_00900.jpg,0
 409 | 6K_eot6tev4_06128.jpg,0
 410 | ee2hF-Yvs0I_08265.jpg,0
 411 | T2HQxDlcaMY_03017.jpg,0
 412 | elausNZXcg0_00859.jpg,0
 413 | a3L1qZp3_DA_09213.jpg,0
 414 | Jku3hUJPpH0_07002.jpg,0
 415 | XJAHgPM1XPI_02419.jpg,0
 416 | qmoqJRP5N6E_04347.jpg,0
 417 | c4EtP3wo-bE_09447.jpg,0
 418 | tF3MjDvzNdc_00087.jpg,0
 419 | KE8X0OdyKDQ_01759.jpg,0
 420 | iaMLqE47p0A_03092.jpg,0
 421 | CqKRF04ppXs_02228.jpg,0
 422 | 6QQiveo7dG8_11105.jpg,0
 423 | zkNq3v3cAvI_03353.jpg,0
 424 | YKXvjIq4wCU_04590.jpg,0
 425 | q_TVIZF2nOE_01499.jpg,2
 426 | 4oMBmMnd50M_08306.jpg,0
 427 | IgQQwQbbsiY_05263.jpg,0
 428 | _U96lwY2IGg_08244.jpg,0
 429 | XgkfqIrDeS4_10616.jpg,0
 430 | gMBlQGY5Tws_00032.jpg,0
 431 | TA3Uo3a9674_02467.jpg,0
 432 | ee2hF-Yvs0I_09523.jpg,0
 433 | l5lCMjVWzZY_03176.jpg,0
 434 | rSN7S4hvlDU_05337.jpg,0
 435 | 5LuVPUmLddw_05852.jpg,137
 436 | RgtdGbpoKpA_07173.jpg,0
 437 | yuOkyxmxtf8_04529.jpg,0
 438 | GlxPOLCc3BY_06268.jpg,0
 439 | sY6kIBMKtt8_04241.jpg,0
 440 | ojATXPBrCYc_01470.jpg,0
 441 | n4L1eV5hQ7s_07769.jpg,0
 442 | yc10Vpsv6_g_03414.jpg,0
 443 | T01wm-Oir3I_01358.jpg,0
 444 | qhbAPfOvCaM_04660.jpg,0
 445 | AgRgbigzBfI_03750.jpg,0
 446 | qNWl-rF7eBQ_05941.jpg,0
 447 | lZQlFR9EgIc_05822.jpg,0
 448 | FAi0uzphqus_04426.jpg,0
 449 | vaWc73AD648_00397.jpg,0
 450 | 1ugesWXaSmQ_01762.jpg,0
 451 | GBZRXWezBHM_00571.jpg,0
 452 | _U96lwY2IGg_07200.jpg,0
 453 | 2cRvAaFHhV4_02581.jpg,0
 454 | tyX4r9UrQ_M_03720.jpg,0
 455 | 5z0e-GKJA10_03796.jpg,0
 456 | 82X5uEFpEVQ_05216.jpg,0
 457 | Z6KavxJTyaM_05307.jpg,0
 458 | SIeYy4k8MWs_04597.jpg,0
 459 | H1WgQPvOJkI_00162.jpg,0
 460 | uOlG-nIP6sI_00107.jpg,0
 461 | rCYEXgDUUuc_07802.jpg,0
 462 | ZaYGprSYRTE_09432.jpg,0
 463 | KOB2sQtrMck_02557.jpg,0
 464 | Jfa6FUzKDRk_12356.jpg,0
 465 | XOUTZXULnsQ_09356.jpg,0
 466 | rr09pDsDb3g_03275.jpg,0
 467 | VSx3l0B3uQA_00000.jpg,158
 468 | vsltvfLV9NY_04227.jpg,158
 469 | d59CUdGg2hs_03993.jpg,158
 470 | F_wXHFab5XU_04468.jpg,158
 471 | g2L0U0MpH44_02329.jpg,158
 472 | k6Kq1HpaeBM_01939.jpg,158
 473 | FmPeW-gaVOM_04409.jpg,128
 474 | 8Dn1hNcZoU0_04985.jpg,158
 475 | OygmSwWfkBE_02546.jpg,158
 476 | QgEeTP_9S5k_04496.jpg,158
 477 | jMFxoLjFJr0_01925.jpg,128
 478 | XD0UKQ4TG8A_03021.jpg,158
 479 | KMz8f9vDK38_00230.jpg,158
 480 | G4rTrxLTNro_02830.jpg,158
 481 | coM4d1CQZfs_05950.jpg,158
 482 | QM2rKnTBDOs_02525.jpg,158
 483 | VaHxrD0HXZk_05336.jpg,158
 484 | O5gntzkB5zY_06990.jpg,158
 485 | JYtSnjy4zAk_05430.jpg,158
 486 | g2LGqKGPcYg_02339.jpg,128
 487 | Srql9_3RP6w_03110.jpg,158
 488 | A6im9EhEI2I_09586.jpg,158
 489 | C-y1gdM_qhs_03724.jpg,158
 490 | pZxdCZUAPjo_06325.jpg,158
 491 | kZMjT8xhRwk_09102.jpg,158
 492 | Dy4w7hVfAHE_00058.jpg,158
 493 | WvaTCLyexsM_06142.jpg,158
 494 | NRSaNlv6apI_01874.jpg,158
 495 | JbnDi-VbwFQ_00656.jpg,76
 496 | iyLpH7UMi4Y_00000.jpg,158
 497 | csGpqnok6p8_00733.jpg,158
 498 | VxsbFCxI2lA_11077.jpg,158
 499 | aRMH5Fu19Ho_04426.jpg,158
 500 | a_WN9y95Q6Y_08521.jpg,158
 501 | Gf_2iIW9Eus_01372.jpg,158
 502 | dQitFgVsZt8_06703.jpg,158
 503 | Gpg5eTbos_M_09765.jpg,158
 504 | RAF_jwuyR4s_01367.jpg,158
 505 | g2L0U0MpH44_06014.jpg,158
 506 | VaHxrD0HXZk_03927.jpg,158
 507 | 8LASrlnL9zE_07340.jpg,158
 508 | wEEfDatopzU_03621.jpg,158
 509 | gr4PJrV-eps_03593.jpg,158
 510 | J_uAPfZA7KE_01108.jpg,158
 511 | Yw0Y4bj935E_01627.jpg,113
 512 | ODRwpu3F4l8_00817.jpg,158
 513 | yDn1TDQXFFw_03266.jpg,158
 514 | EINFmlJGIWw_01435.jpg,158
 515 | iyLpH7UMi4Y_02373.jpg,158
 516 | fAgkHGypXWU_02898.jpg,158
 517 | QdJXjg6OjNA_06930.jpg,158
 518 | Th0zMycdYeo_01704.jpg,158
 519 | v51i1IqiIgs_07686.jpg,158
 520 | 08n8aWx2Pzs_04460.jpg,158
 521 | ef2Dn9Ht4O8_00374.jpg,158
 522 | Mz5hC8y6fzc_05776.jpg,158
 523 | VxsbFCxI2lA_06210.jpg,158
 524 | O5gntzkB5zY_02523.jpg,158
 525 | MV08II8DD3o_01699.jpg,158
 526 | I-WOwyxqcxY_00771.jpg,128
 527 | 6QhhRDHOri4_02901.jpg,158
 528 | yOIYswGsASk_01721.jpg,76
 529 | VSx3l0B3uQA_00166.jpg,158
 530 | x-CmQyPQLrs_06369.jpg,158
 531 | 9iC5ZqBk8Dw_00143.jpg,158
 532 | _osfdlMLggk_02803.jpg,158
 533 | PC1zN1VNvIo_02253.jpg,158
 534 | VaHxrD0HXZk_05784.jpg,158
 535 | 224RvvoNafs_05176.jpg,158
 536 | IK0UTdvSP78_00539.jpg,158
 537 | RAF_jwuyR4s_01676.jpg,158
 538 | iMb-I_r2sZs_01663.jpg,158
 539 | 3wpEAM9W7y4_00000.jpg,98
 540 | iMb-I_r2sZs_09123.jpg,158
 541 | CXWBtIJ6YRQ_03862.jpg,158
 542 | NIwa4C1C9OM_09828.jpg,158
 543 | yJoHQK7jO6g_01477.jpg,158
 544 | S_2VqNXgoA0_00977.jpg,158
 545 | iMb-I_r2sZs_00232.jpg,158
 546 | S7lIOKvu9ik_04776.jpg,128
 547 | YYCnr9SzghQ_00163.jpg,158
 548 | YfjIhCOE4Ng_01475.jpg,158
 549 | yZe8d2C-fzk_02340.jpg,158
 550 | JrvIn57wQ5I_06333.jpg,158
 551 | J6Iksi76MNg_00860.jpg,158
 552 | ip9qfm3bpA4_04345.jpg,158
 553 | JgXUkYCbAgk_05804.jpg,158
 554 | NIwa4C1C9OM_04269.jpg,158
 555 | taPTPxspRQ0_02268.jpg,158
 556 | NIwa4C1C9OM_11708.jpg,158
 557 | WdNYal09Pfw_02020.jpg,158
 558 | YgXzpWAEhto_01859.jpg,158
 559 | kZMjT8xhRwk_12253.jpg,158
 560 | VEUSx17jCXQ_04976.jpg,158
 561 | P8CZCX42X-o_00130.jpg,158
 562 | 1LnpIduVplM_08632.jpg,158
 563 | IOsrEqzStgg_00648.jpg,158
 564 | OgXKdhdrFts_05850.jpg,158
 565 | JbnDi-VbwFQ_01270.jpg,157
 566 | XV7d_zJH7wg_07545.jpg,158
 567 | lIFkAPQOMjw_01301.jpg,158
 568 | vNmcz_wPYr0_04998.jpg,158
 569 | SSuQ0s9gKb8_01400.jpg,158
 570 | zATmT9rLfpc_08349.jpg,158
 571 | wuzTyJ39txE_03994.jpg,158
 572 | sLLh1tM0tFg_03486.jpg,158
 573 | JHAEvkLc-S0_01448.jpg,128
 574 | 6L0MRj18a8E_07942.jpg,128
 575 | WYVKLOi5p5Q_05180.jpg,57
 576 | N_9eucfmvj8_01920.jpg,113
 577 | 9HiLLaJKJF0_02260.jpg,113
 578 | 2snLA3j1Sis_11759.jpg,128
 579 | vuaIJq5YKKA_04348.jpg,113
 580 | bzphhT_NKV0_07496.jpg,113
 581 | 2prWa2jwgpY_00277.jpg,113
 582 | Wq8zx_bIe2o_00763.jpg,158
 583 | OGWaVBdUKM0_00894.jpg,128
 584 | 0D7lD8AGmLY_01804.jpg,52
 585 | OlBWHhugIzI_01985.jpg,128
 586 | UXz3k5umHfQ_00972.jpg,24
 587 | o9xvzfPYzrg_01275.jpg,128
 588 | btWTAD1q_UQ_04639.jpg,113
 589 | fOM0SJrCWjQ_05546.jpg,113
 590 | lcZGhebGgP8_04583.jpg,113
 591 | fpdGLl_sWsQ_04821.jpg,64
 592 | fOM0SJrCWjQ_09319.jpg,113
 593 | G5p4GNFP7oU_03501.jpg,94
 594 | OTcxSn6gL5M_08849.jpg,128
 595 | auK3VtXE68c_01556.jpg,128
 596 | BsceC-igcTY_01370.jpg,113
 597 | N8zUIdGppJE_03225.jpg,113
 598 | ip9qfm3bpA4_00825.jpg,128
 599 | Wzn3FjH1eYc_00804.jpg,113
 600 | lcZGhebGgP8_05456.jpg,128
 601 | Y4EASeZ6IeM_01643.jpg,24
 602 | kcD_xN-LqAU_02926.jpg,113
 603 | as24tyYm4tI_06550.jpg,52
 604 | 3bqbvTgnVR0_01344.jpg,113
 605 | 4ZGxBi_y9XA_01172.jpg,128
 606 | 4Coe0QNTdg0_03185.jpg,113
 607 | POTLb6GMka0_05874.jpg,160
 608 | jOfS1-trrR0_00548.jpg,113
 609 | jO4JPSLPGfc_07662.jpg,158
 610 | wKAZcMTj-Xg_01619.jpg,158
 611 | XyHOQt3r_IQ_02201.jpg,113
 612 | 6L0MRj18a8E_14077.jpg,128
 613 | cq6OVN_7LLQ_03911.jpg,52
 614 | jv4y-IsB8d8_01385.jpg,113
 615 | 26hhv1v9b1U_03427.jpg,52
 616 | PYUIHcnAgbc_05564.jpg,160
 617 | fOM0SJrCWjQ_01892.jpg,24
 618 | VGFnezvBv7U_05583.jpg,0
 619 | 60QrLYe5iXo_05718.jpg,128
 620 | Ytp-pm301Uo_02632.jpg,23
 621 | 02y8nDgo140_02401.jpg,128
 622 | d3y17D_K9Zw_04314.jpg,76
 623 | d3y17D_K9Zw_00152.jpg,52
 624 | 02y8nDgo140_01988.jpg,76
 625 | kHLHD2EX8wk_01490.jpg,76
 626 | 35oa3d_1zy4_02400.jpg,76
 627 | hdlOXYGh608_00800.jpg,76
 628 | JIWgLmN8OQ0_02243.jpg,76
 629 | 10AKUYRjDPk_01136.jpg,76
 630 | V3RTIzd6rm4_02661.jpg,76
 631 | e0gb9mzMNII_01376.jpg,76
 632 | lHZHc0KgUk0_02816.jpg,76
 633 | 9DLmkr7Jqp4_04828.jpg,76
 634 | lHZHc0KgUk0_01905.jpg,76
 635 | d3y17D_K9Zw_07899.jpg,76
 636 | JIWgLmN8OQ0_02411.jpg,76
 637 | 4L5cdskEiA4_04000.jpg,76
 638 | JIWgLmN8OQ0_01964.jpg,76
 639 | jszPZiqEheI_00237.jpg,76
 640 | jszPZiqEheI_06380.jpg,76
 641 | hdlOXYGh608_00365.jpg,76
 642 | kHLHD2EX8wk_00494.jpg,76
 643 | hdlOXYGh608_03270.jpg,76
 644 | 4yy4oy4dyO0_04418.jpg,76
 645 | hdlOXYGh608_00489.jpg,76
 646 | d3y17D_K9Zw_04843.jpg,76
 647 | Px27gP__n5M_01460.jpg,76
 648 | kHLHD2EX8wk_02522.jpg,76
 649 | 4yy4oy4dyO0_00589.jpg,76
 650 | 4yy4oy4dyO0_04480.jpg,76
 651 | 4yy4oy4dyO0_04901.jpg,76
 652 | o3JjUoyX_Ck_02141.jpg,76
 653 | kHLHD2EX8wk_01092.jpg,76
 654 | d3y17D_K9Zw_08059.jpg,76
 655 | a8Lqhfw65j0_02491.jpg,76
 656 | 02y8nDgo140_02709.jpg,76
 657 | EE9hk0C7pRU_00049.jpg,76
 658 | 9DLmkr7Jqp4_02109.jpg,76
 659 | e0gb9mzMNII_00105.jpg,76
 660 | N6M1i0wb5tU_01404.jpg,76
 661 | 4yy4oy4dyO0_04539.jpg,76
 662 | V3RTIzd6rm4_00225.jpg,76
 663 | d3y17D_K9Zw_03228.jpg,76
 664 | kHLHD2EX8wk_00818.jpg,76
 665 | 9DLmkr7Jqp4_03464.jpg,76
 666 | jszPZiqEheI_05683.jpg,76
 667 | a8Lqhfw65j0_06218.jpg,76
 668 | d3y17D_K9Zw_07920.jpg,76
 669 | PmKIY6Qy_D4_00240.jpg,76
 670 | 4yy4oy4dyO0_00611.jpg,76
 671 | Aj93kqCGc-E_01950.jpg,52
 672 | Px27gP__n5M_01251.jpg,76
 673 | PYma9jaszTE_00359.jpg,76
 674 | Y97frE29gn8_01289.jpg,76
 675 | jszPZiqEheI_10296.jpg,76
 676 | 9DLmkr7Jqp4_05847.jpg,76
 677 | Prb4Lgqc-Lg_02614.jpg,76
 678 | lHZHc0KgUk0_11964.jpg,76
 679 | oZDCSNMwbt4_05311.jpg,76
 680 | 9DLmkr7Jqp4_03094.jpg,76
 681 | jszPZiqEheI_00373.jpg,76
 682 | gmquYfr-GxA_00247.jpg,76
 683 | 02y8nDgo140_06769.jpg,76
 684 | 10AKUYRjDPk_01174.jpg,76
 685 | hdlOXYGh608_04006.jpg,76
 686 | a8Lqhfw65j0_06238.jpg,76
 687 | e0gb9mzMNII_00514.jpg,76
 688 | lHZHc0KgUk0_03664.jpg,76
 689 | hdlOXYGh608_03362.jpg,76
 690 | I0UB7Twv36M_02739.jpg,76
 691 | Px27gP__n5M_03309.jpg,76
 692 | e0gb9mzMNII_01724.jpg,26
 693 | NSeCrJtQFCU_03494.jpg,76
 694 | sentS7QjeeA_03110.jpg,76
 695 | d3y17D_K9Zw_07955.jpg,76
 696 | 4yy4oy4dyO0_00573.jpg,76
 697 | Px27gP__n5M_01059.jpg,76
 698 | dSTwxFD9Tkk_00275.jpg,76
 699 | 4yy4oy4dyO0_04996.jpg,76
 700 | GzBY4DsM3r4_01808.jpg,76
 701 | kHLHD2EX8wk_01447.jpg,76
 702 | lHZHc0KgUk0_04317.jpg,76
 703 | Prb4Lgqc-Lg_00280.jpg,76
 704 | IzBXN0p5w4c_01056.jpg,76
 705 | lHZHc0KgUk0_11147.jpg,76
 706 | dSTwxFD9Tkk_01199.jpg,76
 707 | e0gb9mzMNII_01091.jpg,76
 708 | 02y8nDgo140_02463.jpg,76
 709 | I-XAPA1RzW4_07446.jpg,23
 710 | 9DLmkr7Jqp4_05909.jpg,76
 711 | 02y8nDgo140_00283.jpg,76
 712 | jszPZiqEheI_09935.jpg,76
 713 | V3RTIzd6rm4_00690.jpg,76
 714 | 9DLmkr7Jqp4_03707.jpg,76
 715 | PYma9jaszTE_00982.jpg,76
 716 | d3y17D_K9Zw_07964.jpg,76
 717 | d3y17D_K9Zw_03056.jpg,76
 718 | 4yy4oy4dyO0_00641.jpg,76
 719 | NSeCrJtQFCU_04019.jpg,76
 720 | 75RU2OVUdCw_04000.jpg,76
 721 | EE9hk0C7pRU_00508.jpg,76
 722 | JIWgLmN8OQ0_02628.jpg,76
 723 | I-XAPA1RzW4_07720.jpg,76
 724 | oZDCSNMwbt4_03898.jpg,76
 725 | kHLHD2EX8wk_00886.jpg,76
 726 | N6M1i0wb5tU_01979.jpg,76
 727 | dfiWC-JczpA_01997.jpg,76
 728 | 10AKUYRjDPk_01076.jpg,76
 729 | JIWgLmN8OQ0_00585.jpg,76
 730 | hdlOXYGh608_02852.jpg,76
 731 | Prb4Lgqc-Lg_04314.jpg,76
 732 | JIWgLmN8OQ0_00951.jpg,76
 733 | JIWgLmN8OQ0_00377.jpg,76
 734 | LZSqIUF6H6c_00098.jpg,76
 735 | Px27gP__n5M_03347.jpg,76
 736 | 5h8-A9ivnXA_00952.jpg,76
 737 | P5x6TshVD8I_02400.jpg,24
 738 | d3y17D_K9Zw_04807.jpg,76
 739 | lHZHc0KgUk0_10006.jpg,76
 740 | kHLHD2EX8wk_00120.jpg,76
 741 | NHZ9-N_0jxU_01669.jpg,151
 742 | aeLSU9eUxE0_07413.jpg,151
 743 | A5_OnduJ56c_01607.jpg,151
 744 | BJ3uiLhfB8k_01984.jpg,26
 745 | g6GnXbkJ1wE_00934.jpg,151
 746 | aeLSU9eUxE0_07810.jpg,151
 747 | CHUqrPQKmxc_01776.jpg,151
 748 | NHZ9-N_0jxU_06254.jpg,151
 749 | pBHDqmYBTtY_12585.jpg,151
 750 | WkhMjUmweWg_02872.jpg,151
 751 | jzZRvW87cP4_08754.jpg,151
 752 | DYzngc9Fkzc_00317.jpg,151
 753 | 2FpLluGSgeg_00129.jpg,151
 754 | g6GnXbkJ1wE_06753.jpg,151
 755 | KCbc_HsiVuw_00769.jpg,151
 756 | g9pynYe986E_00432.jpg,0
 757 | Q4hi_RzbGFg_00000.jpg,151
 758 | xx1oCgLDYjM_04981.jpg,151
 759 | v-j8vZLZoWI_07996.jpg,151
 760 | vGqbU_i5swY_09663.jpg,0
 761 | 9M0bdzxxtFI_09312.jpg,13
 762 | sQ2IdnJa_Pc_00746.jpg,151
 763 | 9m4i1zJDPdU_03153.jpg,190
 764 | aeLSU9eUxE0_07965.jpg,151
 765 | A5_OnduJ56c_01534.jpg,151
 766 | Knd-CGqCpl4_00000.jpg,151
 767 | nCEVSP34zEQ_00300.jpg,151
 768 | pBHDqmYBTtY_13634.jpg,151
 769 | IPZ6C4CLS_4_00091.jpg,56
 770 | xx1oCgLDYjM_00716.jpg,151
 771 | JdjPkZUb178_02918.jpg,151
 772 | g6GnXbkJ1wE_02816.jpg,151
 773 | aeLSU9eUxE0_01351.jpg,151
 774 | qrqClKzcf-w_00224.jpg,151
 775 | aeLSU9eUxE0_00136.jpg,151
 776 | WkhMjUmweWg_05328.jpg,151
 777 | _XIf0deyzx4_00778.jpg,151
 778 | oY_7UgvaBVA_12657.jpg,151
 779 | 9Eb1UI18PuE_09838.jpg,56
 780 | sQ2IdnJa_Pc_00546.jpg,151
 781 | CHUqrPQKmxc_05766.jpg,151
 782 | WkhMjUmweWg_07092.jpg,151
 783 | CHUqrPQKmxc_08592.jpg,151
 784 | Swr9lMlLS28_00556.jpg,151
 785 | aeLSU9eUxE0_00914.jpg,151
 786 | anpjGctNOlI_00145.jpg,151
 787 | 6DYA8D41Hck_07758.jpg,151
 788 | bCtSvft9dPA_02187.jpg,151
 789 | KCbc_HsiVuw_00716.jpg,151
 790 | 6DYA8D41Hck_08425.jpg,151
 791 | zeEvh2dfVC4_06950.jpg,151
 792 | IIeP7dNkoes_01232.jpg,135
 793 | 0fi2Z6HRUYk_00921.jpg,135
 794 | 4K4gSzzITvc_02778.jpg,135
 795 | hWukFEFUyiM_03223.jpg,135
 796 | th8cvB_wQLc_03689.jpg,0
 797 | WuM-EAnor4I_00677.jpg,135
 798 | 9hGoTnTCyRo_02390.jpg,135
 799 | Gye-vQN7T8s_00483.jpg,135
 800 | 2s-ex4NPlfg_12671.jpg,135
 801 | 6uXjW9tiBXE_03588.jpg,135
 802 | D-xGN5uTZuE_00960.jpg,120
 803 | Ox2AyzuwKq8_00692.jpg,135
 804 | 3lhzVwYrk0I_04568.jpg,135
 805 | 2MBmPqHLgac_01914.jpg,135
 806 | IlC4dfNhPW4_01864.jpg,135
 807 | 2MBmPqHLgac_01091.jpg,135
 808 | IWOGZYB4IEg_00150.jpg,119
 809 | diHY5kXGNZw_04493.jpg,119
 810 | 7I_X57GcSTw_03768.jpg,135
 811 | Ua76jzbmus0_04670.jpg,135
 812 | PZcwQnuke8A_00000.jpg,135
 813 | 2MBmPqHLgac_02242.jpg,135
 814 | uC3HScnl4rk_00349.jpg,135
 815 | Urkmxrqoqrs_00467.jpg,111
 816 | XikWavpuBOU_12745.jpg,135
 817 | 3lhzVwYrk0I_02044.jpg,135
 818 | NXjky8gDWx4_00196.jpg,10
 819 | 1KQ2-gTqFfA_01838.jpg,135
 820 | cBFeyEDpeFw_05434.jpg,135
 821 | 03WYjmi2Wqw_03748.jpg,156
 822 | POMr9tEVlJg_02335.jpg,135
 823 | POMr9tEVlJg_02348.jpg,135
 824 | Pm9cRIhYQvg_00106.jpg,135
 825 | GMgGDgOt4CQ_03543.jpg,135
 826 | 2bLJM05xgnk_05235.jpg,135
 827 | 3lhzVwYrk0I_04585.jpg,135
 828 | 6bT00gb4ABs_00086.jpg,135
 829 | 0xLdTFHiDiQ_01501.jpg,82
 830 | 9uMtkrrNgzQ_08925.jpg,135
 831 | xm2LtVunkBA_02446.jpg,135
 832 | tv3PT3AjbKQ_04297.jpg,135
 833 | GMgGDgOt4CQ_02109.jpg,135
 834 | MCbpodfaEho_01803.jpg,135
 835 | io8BL9CA8pQ_00324.jpg,135
 836 | l0zsKQrYQSI_09000.jpg,135
 837 | 3gKb0Hz5t14_03116.jpg,135
 838 | vY6M-k-wIHg_00888.jpg,135
 839 | Ende6KWg4S8_00635.jpg,135
 840 | BQqyVo0YSD0_07316.jpg,11
 841 | 0xLdTFHiDiQ_01932.jpg,135
 842 | GMgGDgOt4CQ_09044.jpg,135
 843 | MCbpodfaEho_00990.jpg,135
 844 | Knd-CGqCpl4_02807.jpg,135
 845 | KSyy3aAwGjo_04892.jpg,135
 846 | FIuD9k1B0vE_02188.jpg,135
 847 | 4K4gSzzITvc_01955.jpg,135
 848 | KSyy3aAwGjo_05505.jpg,135
 849 | G0TmQJc7ExE_01060.jpg,20
 850 | l-Qr4NcgOzA_01674.jpg,135
 851 | KtQHd1DPNtM_12565.jpg,135
 852 | assosxg_D00_01178.jpg,120
 853 | GMgGDgOt4CQ_03202.jpg,15
 854 | tv3PT3AjbKQ_02141.jpg,135
 855 | rzhw6a5CBWA_10500.jpg,111
 856 | YAO5r0_7xF0_00777.jpg,135
 857 | hWukFEFUyiM_00933.jpg,135
 858 | 2MBmPqHLgac_07643.jpg,135
 859 | 2MBmPqHLgac_00442.jpg,135
 860 | Ox2AyzuwKq8_00970.jpg,135
 861 | BlSPiip_TG4_04405.jpg,135
 862 | KtQHd1DPNtM_16281.jpg,135
 863 | 2MBmPqHLgac_00654.jpg,135
 864 | Nq1PRGStn5w_05009.jpg,135
 865 | KtQHd1DPNtM_17352.jpg,135
 866 | assosxg_D00_03405.jpg,120
 867 | _xYXxYh6J24_09292.jpg,135
 868 | l-Qr4NcgOzA_03945.jpg,135
 869 | x8VTrWgg0ZI_05214.jpg,135
 870 | uCromp-klUU_05112.jpg,135
 871 | 5I8dKRzgk5Q_01160.jpg,135
 872 | 7We744fW76w_06634.jpg,135
 873 | GMgGDgOt4CQ_03363.jpg,135
 874 | l-Qr4NcgOzA_02238.jpg,135
 875 | 44DxEt_t34w_00000.jpg,11
 876 | KSyy3aAwGjo_04680.jpg,135
 877 | KtQHd1DPNtM_07225.jpg,135
 878 | KFP5ZnveEbg_00457.jpg,15
 879 | 5yUR4x7eLV8_15110.jpg,128
 880 | 6uXjW9tiBXE_02685.jpg,135
 881 | pGme0VCAoBQ_00702.jpg,135
 882 | OaUEbwHgXZg_01386.jpg,135
 883 | N8FmPXOHyo4_09193.jpg,135
 884 | OJeOeoZ497I_01025.jpg,135
 885 | KSyy3aAwGjo_04667.jpg,135
 886 | 6sp3K9qyNBo_01013.jpg,135
 887 | CRHJY2md1DU_04692.jpg,135
 888 | kkX5EnH7da4_00685.jpg,120
 889 | G0TmQJc7ExE_00440.jpg,135
 890 | PrU6mkhPqOs_00505.jpg,135
 891 | 0fi2Z6HRUYk_05805.jpg,135
 892 | cBFeyEDpeFw_05410.jpg,135
 893 | KSyy3aAwGjo_04913.jpg,135
 894 | YiBzK_-MT3Y_04950.jpg,135
 895 | 3lhzVwYrk0I_03585.jpg,6
 896 | diHY5kXGNZw_04230.jpg,135
 897 | KtQHd1DPNtM_07153.jpg,135
 898 | GMgGDgOt4CQ_03256.jpg,95
 899 | DUj8fqDwHgM_00184.jpg,120
 900 | Nq1PRGStn5w_04888.jpg,135
 901 | MCbpodfaEho_01515.jpg,135
 902 | KSyy3aAwGjo_02108.jpg,135
 903 | Ende6KWg4S8_00147.jpg,135
 904 | DySK3Yz0Wjo_01570.jpg,0
 905 | KtQHd1DPNtM_14688.jpg,135
 906 | pqUs6DMkEbg_07358.jpg,135
 907 | gbmsYeUpSrU_12757.jpg,135
 908 | KtQHd1DPNtM_16258.jpg,135
 909 | STK51jOsBMg_00820.jpg,135
 910 | BlSPiip_TG4_03333.jpg,135
 911 | 8x-kraZ1MCg_01023.jpg,135
 912 | epwvJAICClA_01974.jpg,135
 913 | mdpxyUAo9UQ_00000.jpg,135
 914 | Y3_NF6VUAww_12751.jpg,135
 915 | 23IKGnbplVE_01096.jpg,11
 916 | hWukFEFUyiM_03449.jpg,135
 917 | tv3PT3AjbKQ_02553.jpg,135
 918 | uVkSKh2gefA_05249.jpg,135
 919 | KtQHd1DPNtM_06671.jpg,135
 920 | 2MBmPqHLgac_06983.jpg,135
 921 | epwvJAICClA_00116.jpg,135
 922 | uCromp-klUU_10362.jpg,135
 923 | io8BL9CA8pQ_00185.jpg,135
 924 | DeU6wn0psrU_10437.jpg,135
 925 | uC3HScnl4rk_00537.jpg,135
 926 | CMCpnJZin1k_01481.jpg,135
 927 | NXjky8gDWx4_02210.jpg,135
 928 | UdYYaJLCDNc_00196.jpg,135
 929 | OaUEbwHgXZg_02815.jpg,135
 930 | 2MBmPqHLgac_00847.jpg,135
 931 | DySK3Yz0Wjo_00206.jpg,135
 932 | Ua76jzbmus0_04761.jpg,135
 933 | qZUKTyRtL0Y_06652.jpg,135
 934 | mu9Nk4BVQAs_00238.jpg,135
 935 | pqUs6DMkEbg_06625.jpg,135
 936 | r7oWUAZzQPg_00640.jpg,173
 937 | CMCpnJZin1k_00960.jpg,135
 938 | 9rwlej8E0sA_04648.jpg,135
 939 | pGme0VCAoBQ_00544.jpg,135
 940 | LQhYbfIz0n4_11605.jpg,135
 941 | pGme0VCAoBQ_02226.jpg,135
 942 | 2MBmPqHLgac_02905.jpg,135
 943 | qs0ivcVGGCc_00257.jpg,135
 944 | xm2LtVunkBA_03146.jpg,135
 945 | DVP6veEDRgE_12015.jpg,135
 946 | KSyy3aAwGjo_05185.jpg,135
 947 | WuM-EAnor4I_00825.jpg,135
 948 | 2MBmPqHLgac_01347.jpg,135
 949 | 2MBmPqHLgac_01257.jpg,135
 950 | epwvJAICClA_02880.jpg,135
 951 | tv3PT3AjbKQ_02061.jpg,135
 952 | 4U2ukuu6NwM_12256.jpg,135
 953 | 2SjB3RXZlJw_06272.jpg,135
 954 | r98c7M6CxFo_00182.jpg,135
 955 | YAO5r0_7xF0_04173.jpg,135
 956 | KSyy3aAwGjo_01154.jpg,135
 957 | bP1UMqJuW1I_00732.jpg,135
 958 | AxC2dQJXVNI_01229.jpg,156
 959 | E8h6KLaYB8g_08837.jpg,156
 960 | HdMy_tP2ygA_05561.jpg,156
 961 | AxC2dQJXVNI_04228.jpg,156
 962 | E8h6KLaYB8g_03800.jpg,156
 963 | bP1UMqJuW1I_00517.jpg,156
 964 | KilJYOKDRNI_00703.jpg,22
 965 | voun26PfIM0_00250.jpg,156
 966 | GfLfR-J8tzQ_03728.jpg,156
 967 | IZ63T51-CXk_01257.jpg,156
 968 | W5piS8JBb3E_00330.jpg,156
 969 | Gb_1hnUUypg_02198.jpg,156
 970 | GiDHqMEY_3A_00118.jpg,119
 971 | Iup3aKRgEgA_01154.jpg,156
 972 | 8IySBl2aq-A_01725.jpg,156
 973 | GfLfR-J8tzQ_00419.jpg,156
 974 | E8h6KLaYB8g_03862.jpg,156
 975 | FLYC-hwm-L4_02253.jpg,156
 976 | GfLfR-J8tzQ_03786.jpg,156
 977 | hCRJCCmSgCU_03417.jpg,52
 978 | Fhb323seOkM_03463.jpg,158
 979 | 7pCYXSRzIV0_05544.jpg,128
 980 | IfdmbAXnouw_08413.jpg,128
 981 | j89BWbH4WHY_01543.jpg,52
 982 | eHXuEazoYyQ_00562.jpg,24
 983 | xuE_jmPUsck_00159.jpg,128
 984 | pz8zbiMeM9E_02609.jpg,158
 985 | 2ODVOVPlQKk_00803.jpg,24
 986 | dx23BoTIJ0s_03500.jpg,171
 987 | ZVA5vs8P28k_04199.jpg,128
 988 | FBXboW_fuuc_01500.jpg,158
 989 | PMT6C7uIf4I_02779.jpg,128
 990 | GNMFuCSVWuE_05999.jpg,128
 991 | zlwDTCEwLO8_05406.jpg,158
 992 | w9yX3S9VdtI_01775.jpg,158
 993 | TEa8sdwCweQ_01219.jpg,57
 994 | 9fNEhLf6Hj8_01089.jpg,128
 995 | IGswwuSWemY_02196.jpg,128
 996 | y7-Jx4EXcVs_00881.jpg,128
 997 | An4ROF8msm4_00768.jpg,128
 998 | mVnKrTLdfyY_03994.jpg,158
 999 | yjysF8LMfU4_02747.jpg,128
1000 | 7mU17Eytd4o_02863.jpg,128
1001 | BeCUThzjEcc_02128.jpg,128
1002 | N6M1i0wb5tU_02623.jpg,76
1003 | 7qoIndAr_nc_08960.jpg,95
1004 | b1U6Nx0N0DU_08906.jpg,128
1005 | IYA8Gh-A2WY_03427.jpg,128
1006 | 4h_GzLX7ofo_09808.jpg,128
1007 | 0D7lD8AGmLY_06827.jpg,128
1008 | wTGdeEfZuUI_06008.jpg,158
1009 | Fhb323seOkM_02157.jpg,158
1010 | j-0Is4pGGPk_07269.jpg,52
1011 | NYvsuGZmCwc_08876.jpg,113
1012 | WIk8-l0_QgY_12565.jpg,128
1013 | 0HDsz3C9IHg_01575.jpg,128
1014 | _DR-CilP73k_01764.jpg,128
1015 | Lqow76TRrsg_04006.jpg,128
1016 | 7hjYdua7g0s_00507.jpg,128
1017 | uX-mwU82iV0_00278.jpg,128
1018 | yA-icaUdEpQ_03045.jpg,128
1019 | 6xepifQMHzU_03541.jpg,113
1020 | 7qoIndAr_nc_09067.jpg,95
1021 | hCRJCCmSgCU_02718.jpg,128
1022 | 3gKb0Hz5t14_03335.jpg,128
1023 | j-0Is4pGGPk_00942.jpg,128
1024 | njRpHLoupmM_00969.jpg,128
1025 | cosWi0n6kgk_05402.jpg,25
1026 | Mq7WfGHHZOc_00319.jpg,52
1027 | t5uI1nl0tGE_01802.jpg,158
1028 | WEUYmaBxB2U_02800.jpg,128
1029 | jGSHtdGsZIA_07297.jpg,128
1030 | uXfRAVtj-1g_06138.jpg,128
1031 | 6vw_s9oZ3cc_00864.jpg,128
1032 | XHZ6JpNIAtA_00412.jpg,2
1033 | jGSHtdGsZIA_06874.jpg,128
1034 | 0Ba3vC-gZSA_04221.jpg,128
1035 | SZ8RyPyljcQ_04234.jpg,195
1036 | 44mOIZPyJlA_02432.jpg,24
1037 | K9CH50rwwl4_05004.jpg,128
1038 | b9Sd_oGeM7s_08831.jpg,128
1039 | ThkCQsJn0G4_01833.jpg,128
1040 | jGSHtdGsZIA_05825.jpg,160
1041 | 5sk_9IdGvsU_01244.jpg,95
1042 | fXXLFeJ-4Gc_00205.jpg,76
1043 | P2R9y_CEooo_00080.jpg,76
1044 | H6vu2bnmQPk_02436.jpg,128
1045 | iknh7jgYHtI_00810.jpg,113
1046 | sw04nOHhGQM_03239.jpg,128
1047 | H059c0Ug9ZQ_02634.jpg,128
1048 | Fhb323seOkM_04408.jpg,158
1049 | U70LWOYKgbg_01059.jpg,128
1050 | U4MXNgLPBWQ_02010.jpg,128
1051 | odPajez3afw_05534.jpg,6
1052 | oZDCSNMwbt4_02884.jpg,128
1053 | 6pFhhmI6Tis_04390.jpg,128
1054 | C9AaeTqBRxs_03723.jpg,113
1055 | 2GBmjnWqwQ0_03433.jpg,158
1056 | kBrKBx-kPCE_00073.jpg,128
1057 | mIOI_3SQD_E_01702.jpg,57
1058 | eHz40iUNIsc_00259.jpg,113
1059 | SO4aPP86YSg_05063.jpg,113
1060 | WeuNWwzw94Y_03135.jpg,128
1061 | _FDRwHKVe_c_03313.jpg,128
1062 | 8yWVNhDpyCY_06464.jpg,128
1063 | AYvP_d9c1ns_06754.jpg,128
1064 | uXfRAVtj-1g_01332.jpg,128
1065 | snek0773PjQ_02552.jpg,128
1066 | 8V0wqY0sPaQ_01141.jpg,195
1067 | 7onlrZZJ06M_07164.jpg,128
1068 | XXZwE3A6-Ds_09720.jpg,158
1069 | hsyKbJUiHQM_00183.jpg,24
1070 | 8t6F8abARao_05556.jpg,128
1071 | 6QhhRDHOri4_05005.jpg,158
1072 | Mq7WfGHHZOc_03035.jpg,128
1073 | QErk8RRy8MU_00000.jpg,42
1074 | ZzwpoO7R3eQ_00090.jpg,24
1075 | jGSHtdGsZIA_06857.jpg,128
1076 | JJC66R-hkeQ_04620.jpg,128
1077 | NyA9DSz8TNo_02801.jpg,95
1078 | LUD12S2scoE_00803.jpg,24
1079 | SP7mlQh9H9U_02818.jpg,165
1080 | XXZwE3A6-Ds_08815.jpg,128
1081 | 6wr2X59MqxE_00613.jpg,128
1082 | OjbjyNb66sQ_00000.jpg,128
1083 | 0nifAVL-CmQ_11400.jpg,135
1084 | EUhClqgbYpg_10057.jpg,128
1085 | yA-icaUdEpQ_00326.jpg,128
1086 | oh0JJcf4z6A_07070.jpg,128
1087 | BhsBSmrTB-U_03366.jpg,128
1088 | xItFimDTz0I_02018.jpg,7
1089 | 6AloN1OuwRY_01745.jpg,158
1090 | tCg66bcmuHM_09293.jpg,128
1091 | Px27gP__n5M_03808.jpg,76
1092 | Q5XydFugEoA_05808.jpg,128
1093 | p-BeIWNfsh4_03417.jpg,128
1094 | 3qhUDigU3js_14924.jpg,113
1095 | ahYEJPgBuZ4_05497.jpg,128
1096 | Mq7WfGHHZOc_03016.jpg,128
1097 | 5sk_9IdGvsU_00000.jpg,128
1098 | axQImJQZ7nk_01338.jpg,41
1099 | IfdmbAXnouw_08859.jpg,128
1100 | h7Z9FhQJStk_01044.jpg,128
1101 | AMKAMvV-RYo_13184.jpg,128
1102 | 3gKb0Hz5t14_03817.jpg,156
1103 | KsoeAVsm87Q_00085.jpg,128
1104 | aGzUNYqFf4k_01888.jpg,128
1105 | 8t6F8abARao_05511.jpg,128
1106 | BhsBSmrTB-U_03353.jpg,128
1107 | ebdujdhFRxg_00567.jpg,17
1108 | 1XN00Gvpdv8_00466.jpg,52
1109 | 6pFhhmI6Tis_02511.jpg,128
1110 | 5vSyyrIv4qg_01212.jpg,128
1111 | Va4dDWkpIjw_09601.jpg,128
1112 | ecB556GtP6M_01300.jpg,128
1113 | mIOI_3SQD_E_02117.jpg,57
1114 | 1a1djYxbmx8_02085.jpg,128
1115 | o7e4SSXrubs_01007.jpg,128
1116 | ecB556GtP6M_00988.jpg,128
1117 | xXKeCZSrSaI_00232.jpg,128
1118 | cf-E8KVG-Fk_00859.jpg,128
1119 | 5rJ85zlGf0g_11458.jpg,38
1120 | MepcRUWmsqY_00637.jpg,4
1121 | XCNFA90H38Q_10664.jpg,0
1122 | SqoGHyvSmwA_01109.jpg,52
1123 | 5sk_9IdGvsU_00888.jpg,128
1124 | eA5PyyioV0I_00434.jpg,128
1125 | CTUG73Hw1is_00688.jpg,156
1126 | 5sk_9IdGvsU_00714.jpg,128
1127 | F5rqYBlAdis_03859.jpg,128
1128 | SKnrXEQ3D4U_09335.jpg,128
1129 | 6u5UhvQQWCg_03536.jpg,128
1130 | b2_vvWsNF4o_01034.jpg,128
1131 | M3oluzdz8Tc_01334.jpg,128
1132 | w9yX3S9VdtI_03719.jpg,128
1133 | rcB23aBC5Kw_00647.jpg,64
1134 | WeuNWwzw94Y_02747.jpg,4
1135 | Px27gP__n5M_02208.jpg,128
1136 | 6bHlRlaW3T8_01579.jpg,7
1137 | oh0JJcf4z6A_04912.jpg,128
1138 | E9OpcPgczWU_00598.jpg,128
1139 | GoSlOW9nam4_00088.jpg,128
1140 | pfKZIlrJziY_00491.jpg,7
1141 | 7td2Nt-Cvn8_06001.jpg,128
1142 | 8yWVNhDpyCY_02655.jpg,128
1143 | qMLqr1DB5pU_00236.jpg,128
1144 | 3qhUDigU3js_08082.jpg,128
1145 | CdS6txYuVXc_10451.jpg,128
1146 | 6u5UhvQQWCg_00973.jpg,128
1147 | j89BWbH4WHY_01534.jpg,52
1148 | fEBG6tCTfmY_00999.jpg,113
1149 | EUhClqgbYpg_03519.jpg,128
1150 | 9VV2AshJoJ8_00472.jpg,128
1151 | 0fHwtZ0JsUs_00000.jpg,128
1152 | AYvP_d9c1ns_06692.jpg,128
1153 | 0nifAVL-CmQ_03287.jpg,10
1154 | dI3Lcfl4Fko_00344.jpg,2
1155 | vRtRGAAq9wA_01054.jpg,2
1156 | KLfbfRQgs5w_05986.jpg,100
1157 | 4rZvvkqnatc_07659.jpg,2
1158 | 1ESiE0wuGPQ_01222.jpg,2
1159 | En_tBwfm4d8_07500.jpg,2
1160 | iHbwksas6VA_00000.jpg,2
1161 | gV4TWLH1GAM_11590.jpg,2
1162 | kpeVZI3jbPQ_04250.jpg,2
1163 | vp_9Y7Wq9qQ_00295.jpg,2
1164 | KUsljRxI7cc_00872.jpg,2
1165 | GgaM5RjoddQ_03884.jpg,2
1166 | pg8t7IoypyA_02703.jpg,2
1167 | TbeLumQf9hw_02823.jpg,2
1168 | eJjxkkHCXbk_01077.jpg,2
1169 | 4S8SkQi_Ftc_00174.jpg,0
1170 | A5WaFme4oRk_00599.jpg,2
1171 | yhHVsVKuUeU_12476.jpg,95
1172 | gV4TWLH1GAM_08945.jpg,2
1173 | kpeVZI3jbPQ_06333.jpg,2
1174 | wTE-1K0lppc_01032.jpg,2
1175 | 4rZvvkqnatc_07575.jpg,2
1176 | EyLr-tomU20_07962.jpg,95
1177 | GgaM5RjoddQ_08581.jpg,2
1178 | kJr-lRnQ9tE_06371.jpg,2
1179 | A5WaFme4oRk_01498.jpg,2
1180 | 1EDsz5b0Q0U_03022.jpg,2
1181 | A5WaFme4oRk_04296.jpg,2
1182 | iZ5o34oK0FI_01080.jpg,0
1183 | kpeVZI3jbPQ_03746.jpg,2
1184 | a343nSIWG4U_00873.jpg,2
1185 | oDMKJJTpEeA_00442.jpg,2
1186 | 4Jo5LFthCkQ_00269.jpg,2
1187 | 9YBYZvpeqUI_02552.jpg,2
1188 | KqcWVJ7l47g_04276.jpg,0
1189 | kpeVZI3jbPQ_02478.jpg,2
1190 | M_H5gk91Ghg_01930.jpg,120
1191 | A5WaFme4oRk_02830.jpg,2
1192 | slJDExeSZCE_10534.jpg,2
1193 | CMxasSA-WwI_02961.jpg,2
1194 | R7HzkNNvS9Y_01582.jpg,95
1195 | fkh1qcvS2LE_00194.jpg,2
1196 | eCjnU_KRaJs_04214.jpg,0
1197 | lRMcBlAVW4I_06756.jpg,2
1198 | aRzKMBG3fSA_00328.jpg,0
1199 | vRtRGAAq9wA_00464.jpg,38
1200 | kpeVZI3jbPQ_02839.jpg,2
1201 | fyp2CZEM_uc_05434.jpg,2
1202 | slJDExeSZCE_02269.jpg,2
1203 | kAOAdI5KccI_00179.jpg,9
1204 | m-cYRoACvmw_00000.jpg,183
1205 | A5WaFme4oRk_01898.jpg,2
1206 | ryo3bhnTePk_00165.jpg,2
1207 | gvOAl5-Bv9c_04126.jpg,2
1208 | oDMKJJTpEeA_02511.jpg,2
1209 | wTE-1K0lppc_01349.jpg,2
1210 | GDY86CvsijY_03156.jpg,0
1211 | 4bx_lxqWukk_01768.jpg,2
1212 | npWJ36zPKXw_02944.jpg,38
1213 | 4rZvvkqnatc_07437.jpg,2
1214 | QNCYEnDNh0w_01113.jpg,2
1215 | dI3Lcfl4Fko_00173.jpg,2
1216 | AFijMWQU7T4_00341.jpg,26
1217 | SZUHNDrKfPQ_04279.jpg,2
1218 | swOWbLHZDAU_01781.jpg,2
1219 | 5K-KeMlp6wg_11717.jpg,2
1220 | GGZ4vZAT0fI_03199.jpg,2
1221 | s5GLUDHmK5k_02836.jpg,2
1222 | gvOAl5-Bv9c_00997.jpg,2
1223 | kpeVZI3jbPQ_01585.jpg,2
1224 | mxBHL7cdn6Q_01164.jpg,2
1225 | uV8MnnemWFQ_00299.jpg,0
1226 | slJDExeSZCE_11030.jpg,2
1227 | KFP5ZnveEbg_02221.jpg,2
1228 | I8_FMCEBHoQ_12887.jpg,95
1229 | iZ5o34oK0FI_01430.jpg,0
1230 | slJDExeSZCE_10492.jpg,2
1231 | Qab_cB058LY_00608.jpg,0
1232 | xaJf6XLu57Q_00376.jpg,0
1233 | 0R00jL4sfYw_04176.jpg,2
1234 | SZUHNDrKfPQ_01167.jpg,0
1235 | gV4TWLH1GAM_10447.jpg,2
1236 | vcSzZSV96rA_00721.jpg,2
1237 | ESaaZ2IcXuU_01569.jpg,2
1238 | pFDXD-9DBlg_01828.jpg,2
1239 | 4bx_lxqWukk_01205.jpg,2
1240 | _S7p-Fhaddw_00528.jpg,2
1241 | gvOAl5-Bv9c_03880.jpg,2
1242 | MEr_Cuf7aB4_00536.jpg,0
1243 | ox6YzyVQww0_01164.jpg,0
1244 | 9H8DcKcA_J4_10007.jpg,2
1245 | TDhrKlcXu_k_01065.jpg,2
1246 | xjXjTjSWPjI_00299.jpg,2
1247 | ESaaZ2IcXuU_03361.jpg,2
1248 | q3i7j1dDMxQ_01556.jpg,2
1249 | GGZ4vZAT0fI_02646.jpg,2
1250 | TDhrKlcXu_k_01115.jpg,95
1251 | nbd-83NP6OQ_00050.jpg,2
1252 | y6QDhc9JzVw_05110.jpg,0
1253 | aA6evhyvis4_10800.jpg,95
1254 | slJDExeSZCE_02470.jpg,2
1255 | KqcWVJ7l47g_03114.jpg,2
1256 | 6ua4d3DSPRc_01515.jpg,2
1257 | dI3Lcfl4Fko_00577.jpg,95
1258 | 694-bJL4d68_00442.jpg,2
1259 | A4svqsJ7U20_02628.jpg,24
1260 | swOWbLHZDAU_02457.jpg,2
1261 | 4rZvvkqnatc_07692.jpg,2
1262 | uV8MnnemWFQ_01993.jpg,2
1263 | 4rZvvkqnatc_02800.jpg,2
1264 | QNCYEnDNh0w_01126.jpg,2
1265 | AtyJ-RY5djk_02165.jpg,128
1266 | sf7goRy6HNw_08667.jpg,135
1267 | aRzKMBG3fSA_01628.jpg,0
1268 | i7kOOpYlLUk_03970.jpg,2
1269 | A5WaFme4oRk_03382.jpg,2
1270 | PrU6mkhPqOs_12447.jpg,2
1271 | En_tBwfm4d8_07240.jpg,2
1272 | GgaM5RjoddQ_03602.jpg,2
1273 | a343nSIWG4U_01059.jpg,2
1274 | 4rZvvkqnatc_07394.jpg,2
1275 | RPbCTMF5L4A_06446.jpg,120
1276 | lbOI0fASw3w_02549.jpg,10
1277 | v7RZMVZ9cUU_14463.jpg,135
1278 | wAXJmUqlnUw_04412.jpg,10
1279 | phXXvbSLidU_03187.jpg,135
1280 | UFU8K7JzwhM_02590.jpg,135
1281 | k8P9reNZ8Es_01624.jpg,8
1282 | dUx08gvy0zw_08707.jpg,10
1283 | v0K3u_NEZAY_01780.jpg,57
1284 | vCw50rKnj5s_01578.jpg,11
1285 | 2x4p5dW8vfE_00606.jpg,119
1286 | dCi4KYLuarY_03185.jpg,128
1287 | LQhYbfIz0n4_01795.jpg,135
1288 | 5I8dKRzgk5Q_01317.jpg,15
1289 | uWi1H8Fgrac_00112.jpg,11
1290 | xT4lXgmniVU_03996.jpg,10
1291 | wAXJmUqlnUw_04408.jpg,10
1292 | xT4lXgmniVU_02981.jpg,10
1293 | dUx08gvy0zw_09434.jpg,27
1294 | dUx08gvy0zw_05685.jpg,10
1295 | 3zs0oKPiXt0_00137.jpg,7
1296 | GNh-SDt3Kz8_02259.jpg,135
1297 | RPbCTMF5L4A_06714.jpg,128
1298 | dyjkVexWrmQ_07343.jpg,10
1299 | S-FbXFynCPQ_05833.jpg,10
1300 | 2x4p5dW8vfE_04808.jpg,135
1301 | 1sM0MRngZ0k_01513.jpg,10
1302 | WuM-EAnor4I_00063.jpg,135
1303 | WuM-EAnor4I_00000.jpg,89
1304 | NVIgqAbZJJc_10947.jpg,135
1305 | QEKp6HlUpDQ_00064.jpg,120
1306 | TYq-NT0PRuQ_08820.jpg,120
1307 | V95V9bnZ4Ow_00514.jpg,128
1308 | QYiSMhGwxcc_01733.jpg,120
1309 | ipSo2tgKJ0Y_00747.jpg,135
1310 | PD0T-Wfuy-E_03515.jpg,120
1311 | ipSo2tgKJ0Y_01417.jpg,135
1312 | icBKhYosqLo_05755.jpg,120
1313 | lchdtYUF4LQ_00930.jpg,135
1314 | k6sYBWBCCQM_00343.jpg,120
1315 | PD0T-Wfuy-E_07351.jpg,120
1316 | sFxjkqzjP8w_06355.jpg,120
1317 | qr4zzMrtH4U_00262.jpg,120
1318 | 0-hzMzEaVc8_02807.jpg,135
1319 | y4yZelVCE3o_02384.jpg,120
1320 | QEKp6HlUpDQ_00234.jpg,120
1321 | WuK9J3EJjkQ_00000.jpg,0
1322 | QYiSMhGwxcc_00000.jpg,120
1323 | HSnIVCo5MXo_02935.jpg,135
1324 | TYq-NT0PRuQ_03297.jpg,120
1325 | 5ldJ9YWC9sc_00979.jpg,135
1326 | RBLep7YHlC0_05103.jpg,120
1327 | jje6ct0nRQs_00132.jpg,0
1328 | icBKhYosqLo_01180.jpg,120
1329 | PD0T-Wfuy-E_07143.jpg,120
1330 | mZrsNs_7B7g_07866.jpg,120
1331 | 2BsSQNWBeok_08489.jpg,120
1332 | icBKhYosqLo_05263.jpg,120
1333 | GiDHqMEY_3A_02129.jpg,135
1334 | 7-cBBHiCmGw_05146.jpg,120
1335 | jvJ8j2vkzKo_02156.jpg,135
1336 | RBLep7YHlC0_05219.jpg,120
1337 | NDyOP2sm7EM_00162.jpg,120
1338 | Fu4DzQ0n4Vw_01508.jpg,120
1339 | Fu4DzQ0n4Vw_02697.jpg,120
1340 | Fu4DzQ0n4Vw_02605.jpg,120
1341 | 2SjB3RXZlJw_05040.jpg,120
1342 | PD0T-Wfuy-E_01970.jpg,120
1343 | PEnqBo4ODSw_00849.jpg,120
1344 | PD0T-Wfuy-E_03585.jpg,120
1345 | RBLep7YHlC0_05900.jpg,120
1346 | Fu4DzQ0n4Vw_03629.jpg,120
1347 | Dxc-3DigTpE_04363.jpg,120
1348 | 2BsSQNWBeok_01588.jpg,120
1349 | 45LbnlaJvtI_03581.jpg,135
1350 | qaKoc6a4hJk_04156.jpg,26
1351 | b26JPrIWjTk_00027.jpg,2
1352 | liyDTcGQ2oQ_02784.jpg,26
1353 | rh6zg74dIBQ_00601.jpg,26
1354 | oVEK1745JGM_03507.jpg,26
1355 | EYEGZw9mZ4g_05066.jpg,26
1356 | cUnf-PiKfq0_03117.jpg,2
1357 | u4lQIh1P53Q_15382.jpg,26
1358 | L5uFAVS9Cdc_02840.jpg,26
1359 | liyDTcGQ2oQ_02472.jpg,26
1360 | CC-XXwpK-Bg_00148.jpg,0
1361 | bRC9dEPdl1g_10145.jpg,26
1362 | PrU6mkhPqOs_07833.jpg,2
1363 | FpJb0msuN1M_05437.jpg,95
1364 | liyDTcGQ2oQ_03894.jpg,26
1365 | u4lQIh1P53Q_15394.jpg,26
1366 | EYEGZw9mZ4g_13392.jpg,26
1367 | liyDTcGQ2oQ_00903.jpg,26
1368 | 1ajVuSVQ61Q_02648.jpg,26
1369 | Dp0Bt2cbcc8_00869.jpg,94
1370 | 5ddCpLubh_4_01136.jpg,94
1371 | QlZPEDh5p_A_00378.jpg,94
1372 | 4EadLrnRhH0_05157.jpg,94
1373 | pADINQsCqVo_11463.jpg,94
1374 | 4EadLrnRhH0_04832.jpg,94
1375 | EuYq2-Rojn4_03727.jpg,94
1376 | 7abf2Vyesrs_02770.jpg,94
1377 | 5P-9RehItu0_01224.jpg,94
1378 | pMnuquRQdyk_02923.jpg,94
1379 | 6dK_L0dG-Pw_00697.jpg,94
1380 | 5ddCpLubh_4_00591.jpg,82
1381 | fBwZWS46_dM_03500.jpg,94
1382 | 6nXepeuhjwA_05654.jpg,94
1383 | E9c7ExiRn7g_00354.jpg,94
1384 | pADINQsCqVo_12727.jpg,94
1385 | 22TRaxYqhRs_05073.jpg,94
1386 | 4EadLrnRhH0_04216.jpg,94
1387 | uZvxqkAJwhU_08987.jpg,94
1388 | OvRdbTQu9HY_14107.jpg,94
1389 | TJR7vXpY3aA_03205.jpg,94
1390 | 4EadLrnRhH0_01108.jpg,94
1391 | CJqsqMf1vE0_01126.jpg,94
1392 | nJkDxm_keOE_00407.jpg,18
1393 | QlZPEDh5p_A_06867.jpg,94
1394 | 4vS8U1fCuHM_02426.jpg,94
1395 | mB1ZWIzoHfo_00244.jpg,94
1396 | 7ObXtDLIzUU_00244.jpg,94
1397 | 22TRaxYqhRs_05167.jpg,94
1398 | 50aax6tq2Bg_01016.jpg,94
1399 | pADINQsCqVo_08890.jpg,94
1400 | 4EadLrnRhH0_06063.jpg,94
1401 | mu31o9zzHQs_01312.jpg,94
1402 | hg4_xdkNvhs_01651.jpg,94
1403 | uzguYX_3q-U_09235.jpg,94
1404 | S3uNZPdV8c8_00000.jpg,82
1405 | 7abf2Vyesrs_03476.jpg,94
1406 | fBwZWS46_dM_01024.jpg,94
1407 | SRsXL7HVSfg_01254.jpg,94
1408 | uyQ2JKbBQT8_03263.jpg,94
1409 | pMnuquRQdyk_00383.jpg,94
1410 | 22TRaxYqhRs_00000.jpg,94
1411 | OmuOEouZdrc_01531.jpg,94
1412 | ygiE53ukOHQ_00460.jpg,94
1413 | nJkDxm_keOE_00641.jpg,94
1414 | pADINQsCqVo_08913.jpg,94
1415 | pADINQsCqVo_12770.jpg,94
1416 | vBz-_6bxiEQ_00623.jpg,120
1417 | pADINQsCqVo_08976.jpg,94
1418 | uyQ2JKbBQT8_05520.jpg,94
1419 | MQEhwzAdec4_00247.jpg,94
1420 | TjboLAnkXpw_02211.jpg,94
1421 | ygiE53ukOHQ_00211.jpg,94
1422 | nEe1Qv-MGYw_03096.jpg,9
1423 | BrdCAlGPKts_03094.jpg,7
1424 | S3uNZPdV8c8_00500.jpg,82
1425 | S3uNZPdV8c8_00821.jpg,82
1426 | Px9Pi1DKTrE_05821.jpg,94
1427 | pADINQsCqVo_04653.jpg,94
1428 | 4EadLrnRhH0_04672.jpg,94
1429 | qlsCC4y77WA_15462.jpg,94
1430 | pADINQsCqVo_08333.jpg,94
1431 | 4IfF6PyV_i4_03113.jpg,94
1432 | HMABgIHLcWs_00527.jpg,94
1433 | 6jHElpHvlC0_00499.jpg,94
1434 | pADINQsCqVo_04117.jpg,94
1435 | pADINQsCqVo_08689.jpg,94
1436 | zrCmHuHBwfs_05763.jpg,127
1437 | uZvxqkAJwhU_10244.jpg,94
1438 | Dp0Bt2cbcc8_01400.jpg,94
1439 | Dp0Bt2cbcc8_00718.jpg,94
1440 | QlZPEDh5p_A_07703.jpg,94
1441 | pADINQsCqVo_08304.jpg,94
1442 | 8JJeDZYbSXQ_01254.jpg,94
1443 | 4EadLrnRhH0_01471.jpg,94
1444 | TjboLAnkXpw_05315.jpg,94
1445 | ofjpgeZUop8_08970.jpg,94
1446 | Gp_rzfOSNPY_02016.jpg,2
1447 | Px9Pi1DKTrE_02288.jpg,94
1448 | pADINQsCqVo_08483.jpg,94
1449 | uyQ2JKbBQT8_03050.jpg,94
1450 | JWQrWHxyikw_01083.jpg,94
1451 | 9TJVA2lwW4A_04551.jpg,94
1452 | FWqjC6BeWN0_00916.jpg,56
1453 | MgLmarTfXs0_00174.jpg,38
1454 | Vb_aBuQKjZ4_08120.jpg,56
1455 | UPwKs00bw5Y_02097.jpg,56
1456 | IlC4dfNhPW4_03774.jpg,2
1457 | 0GNNFBrRz1E_11313.jpg,56
1458 | M5w6-ZoQi0A_01318.jpg,56
1459 | Vb_aBuQKjZ4_00135.jpg,56
1460 | 0GNNFBrRz1E_01572.jpg,56
1461 | IDeKQmR3MCs_01088.jpg,56
1462 | FHYwnl_I4ks_01435.jpg,56
1463 | 9Eb1UI18PuE_04831.jpg,56
1464 | eBdMd8UaT_Y_07965.jpg,56
1465 | st6u8U8l0G4_04079.jpg,56
1466 | QnxuZ74J3Q4_02758.jpg,151
1467 | _ps9MB8pvts_00464.jpg,151
1468 | UPwKs00bw5Y_02784.jpg,56
1469 | cZIW4PPWm8U_09839.jpg,56
1470 | e0bWU7iI8ng_04044.jpg,56
1471 | 0GNNFBrRz1E_12309.jpg,56
1472 | 7BywPb67huY_11957.jpg,56
1473 | e0bWU7iI8ng_02312.jpg,56
1474 | CMCpnJZin1k_02187.jpg,151
1475 | _QWJa0yC7oQ_01381.jpg,56
1476 | qrjg3MLrbYM_01331.jpg,56
1477 | e0bWU7iI8ng_03951.jpg,56
1478 | lkI-7CIJ8DU_01236.jpg,151
1479 | opXy-YBRq7k_01826.jpg,56
1480 | 00sMAa9Noh8_02494.jpg,56
1481 | 8nnD3EJs1rU_02135.jpg,56
1482 | lyj-5x5RDyc_03248.jpg,160
1483 | e0bWU7iI8ng_06148.jpg,56
1484 | aE1ULHa7ViM_04269.jpg,56
1485 | 2s-ex4NPlfg_01627.jpg,41
1486 | qrjg3MLrbYM_01287.jpg,56
1487 | WqspwZorLGU_01734.jpg,56
1488 | Vb_aBuQKjZ4_04548.jpg,56
1489 | FHYwnl_I4ks_03113.jpg,56
1490 | Vb_aBuQKjZ4_03447.jpg,56
1491 | M1d_puqCq3A_00344.jpg,56
1492 | e0bWU7iI8ng_04555.jpg,56
1493 | OeVYRjMYjWY_00258.jpg,56
1494 | dQpPEudltI8_00833.jpg,56
1495 | 0GNNFBrRz1E_02515.jpg,56
1496 | 6-iywj5Foxk_00515.jpg,1
1497 | 5pYB0yx8mH4_01270.jpg,1
1498 | _hVWIKfTSKw_03435.jpg,1
1499 | SlxNn215pkY_05273.jpg,1
1500 | _hVWIKfTSKw_01241.jpg,1
1501 | IQin3x3XHJI_04996.jpg,1
1502 | IQin3x3XHJI_03392.jpg,1
1503 | IQin3x3XHJI_01326.jpg,1
1504 | 0E1PQJg49VU_02888.jpg,1
1505 | XaBPW6n4Nmw_00553.jpg,1
1506 | BSu6OJ5te-I_00670.jpg,114
1507 | ea50_kuzHxE_00486.jpg,1
1508 | joLmTFwxSkQ_03793.jpg,1
1509 | vjTPNqHHmxs_02426.jpg,1
1510 | X4a2pmYx7qk_00990.jpg,1
1511 | 1Y9qmN8MUy0_00029.jpg,1
1512 | 0E1PQJg49VU_06061.jpg,1
1513 | zKqnlxdaHmY_00830.jpg,1
1514 | G9eqsEgNklQ_04767.jpg,1
1515 | CPQbWkWUQsc_01272.jpg,1
1516 | lCbaLww4sFg_01972.jpg,8
1517 | IQin3x3XHJI_05831.jpg,1
1518 | 0yf6y3XBLQ0_00897.jpg,1
1519 | AGgT3ux6B5o_08266.jpg,1
1520 | uGpA4Dl267k_06236.jpg,1
1521 | 1LK0uZkcJZM_02040.jpg,1
1522 | uGpA4Dl267k_02204.jpg,1
1523 | bLaEvly5YKE_01400.jpg,1
1524 | 5jTBpIiFcqQ_00529.jpg,1
1525 | SlxNn215pkY_05159.jpg,1
1526 | oGDHNZOYmbs_13969.jpg,1
1527 | 6-iywj5Foxk_00961.jpg,1
1528 | _hVWIKfTSKw_06080.jpg,1
1529 | qPbNs3S0a8I_05703.jpg,8
1530 | zKqnlxdaHmY_00348.jpg,1
1531 | 15k-5iMko8w_00029.jpg,1
1532 | BSu6OJ5te-I_00585.jpg,114
1533 | IQin3x3XHJI_00617.jpg,1
1534 | 1Y9qmN8MUy0_02693.jpg,1
1535 | m1xUZh0RSdU_02895.jpg,1
1536 | G9eqsEgNklQ_05863.jpg,1
1537 | 2nrZrvYQut0_00676.jpg,1
1538 | vjTPNqHHmxs_01862.jpg,1
1539 | S_JuVX51R2Q_04877.jpg,1
1540 | 1LK0uZkcJZM_00000.jpg,1
1541 | urECG0A42ng_01562.jpg,1
1542 | IQin3x3XHJI_00855.jpg,1
1543 | 5pYB0yx8mH4_00079.jpg,1
1544 | 2K5cUVxXZ54_00530.jpg,1
1545 | 0yf6y3XBLQ0_00186.jpg,19
1546 | zY3yIytHmj0_06518.jpg,1
1547 | _hVWIKfTSKw_06026.jpg,1
1548 | 0b1_1vE8eoA_05949.jpg,1
1549 | _hVWIKfTSKw_04872.jpg,1
1550 | AGgT3ux6B5o_00970.jpg,1
1551 | 7FVLSmCBJD8_11216.jpg,6
1552 | 15k-5iMko8w_02354.jpg,1
1553 | 15k-5iMko8w_02512.jpg,1
1554 | gEQUWj6GNyk_06088.jpg,1
1555 | ea50_kuzHxE_00488.jpg,1
1556 | 1Y9qmN8MUy0_06339.jpg,1
1557 | XYWH_lb76KE_00690.jpg,1
1558 | 2K5cUVxXZ54_00499.jpg,1
1559 | 3d1lsIG-GlI_05586.jpg,156
1560 | lX3Dy2ffqJ0_05795.jpg,24
1561 | Z23iMysaPwQ_03134.jpg,11
1562 | Pq_p5dyb-oY_01056.jpg,11
1563 | M7wRpR4LU8Q_00391.jpg,25
1564 | LW_njAFAhJA_05105.jpg,25
1565 | tQRlZw4Rad0_01926.jpg,24
1566 | M7wRpR4LU8Q_04203.jpg,25
1567 | LiPTQx0ac8Y_01673.jpg,128
1568 | MX9mkBzsMpY_01562.jpg,25
1569 | kduaFZz9oQY_05235.jpg,25
1570 | sCJHZSb7hlA_03312.jpg,25
1571 | nTWOcSS6TbQ_01587.jpg,18
1572 | S_M5s0vIcEM_06109.jpg,94
1573 | pBnlP1qC6no_03840.jpg,128
1574 | kduaFZz9oQY_02690.jpg,25
1575 | 5nb-5pVHydU_00497.jpg,25
1576 | d4cumyUn0HY_04751.jpg,25
1577 | _hau617Vex8_01900.jpg,127
1578 | ExXt6dZ_rNk_00303.jpg,168
1579 | ejjO6ITECow_00213.jpg,168
1580 | t-tnk1LPhnU_00000.jpg,168
1581 | 1Lnw_v-VUPs_02364.jpg,168
1582 | B2mYbfg7jaM_10000.jpg,168
1583 | 1Lnw_v-VUPs_01761.jpg,168
1584 | k4ACuU--Raw_01031.jpg,150
1585 | 1_MROIfd6II_00000.jpg,168
1586 | 1Lnw_v-VUPs_00957.jpg,168
1587 | 1Lnw_v-VUPs_01656.jpg,168
1588 | 1Lnw_v-VUPs_00428.jpg,168
1589 | Atu3oyZlp7o_01000.jpg,135
1590 | JGQNIPyjOuo_00000.jpg,15
1591 | fFkgT3xRgbA_00000.jpg,52
1592 | CzIv55ElPSQ_00000.jpg,128
1593 | RVkriPSwA3Y_05523.jpg,5
1594 | 1-8F-gAFBcs_01573.jpg,204
1595 | EFghGR7xWDI_03699.jpg,204
1596 | be-4UDBsFzU_05071.jpg,2
1597 | VpNu2pQJKfQ_01912.jpg,204
1598 | ossqKiAP0_I_03736.jpg,204
1599 | 39vI5bLSmvA_00530.jpg,95
1600 | ZK1zIc48Q5w_02257.jpg,204
1601 | Ehh23-V3ciQ_06206.jpg,204
1602 | nZbhRK-FrB0_09197.jpg,204
1603 | XBfEnfFbl6o_05676.jpg,204
1604 | rqX_XqeOYBY_00000.jpg,38
1605 | MrpOBBPGmgs_03546.jpg,204
1606 | CB2lyAAS6Mc_03393.jpg,204
1607 | BtXNamdT8rM_08392.jpg,66
1608 | uijYQ07vVh0_04258.jpg,204
1609 | D8AqzVOvU2Y_00842.jpg,120
1610 | Qsp62aoF5Jc_01943.jpg,0
1611 | nip_gkEkegY_01199.jpg,0
1612 | I7h1-C-YdOI_09386.jpg,56
1613 | E36DS7FFTSc_01124.jpg,30
1614 | WX13DJE1wBU_04344.jpg,11
1615 | 6_sLF3OBtc8_03092.jpg,7
1616 | y6Od1Rmtqf8_00140.jpg,11
1617 | lCkGFvqJ-X4_02153.jpg,8
1618 | -8fEeueL6a0_01326.jpg,11
1619 | kdvRSwYAQVA_00439.jpg,11
1620 | BPbizq5KpgQ_01627.jpg,7
1621 | OVbPr2z5IWk_00000.jpg,11
1622 | SiipTqRBde4_04717.jpg,11
1623 | G-9NcHykLTA_04614.jpg,11
1624 | GiiUVJybi7A_09299.jpg,7
1625 | OVbPr2z5IWk_01503.jpg,11
1626 | Hp2S1USUq3U_01580.jpg,11
1627 | tWVZgp-eQG8_01003.jpg,25
1628 | ojhKtzwJTm4_02560.jpg,11
1629 | roiP360wZKE_09850.jpg,11
1630 | x0G08lidQNg_01856.jpg,135
1631 | rYKtWL0bkkg_06231.jpg,11
1632 | HfNlRgtxoT0_06173.jpg,11
1633 | vl0i08iGC2I_00429.jpg,11
1634 | qCTyPUNjADQ_02090.jpg,10
1635 | kpkUswO_HeA_03627.jpg,24
1636 | LX2XnSIF7Qo_06173.jpg,7
1637 | CMxV1Tnb5mg_10460.jpg,7
1638 | ccKsqZj3Vrg_02117.jpg,94
1639 | Xw_BHmByjAM_00202.jpg,11
1640 | yLRXBlXVQRA_00742.jpg,135
1641 | 8z8Nileuq5Y_02481.jpg,11
1642 | -KmsoByCgTE_03894.jpg,44
1643 | o2dQROh_BpA_01617.jpg,1
1644 | NDk9_Oyw1PQ_05260.jpg,11
1645 | y5x9xJ7rLmI_05544.jpg,18
1646 | RO_Myj1_6_M_01525.jpg,94
1647 | r7Gwaoru_gg_00655.jpg,128
1648 | TnUk_-gMnB8_00928.jpg,11
1649 | VPL59Jod_CU_04274.jpg,6
1650 | qZUKTyRtL0Y_02330.jpg,11
1651 | eygquDTqRHo_02951.jpg,11
1652 | QaLmTRpFe3Y_11992.jpg,44
1653 | G3OZvsFdoF0_05911.jpg,24
1654 | _hau617Vex8_05123.jpg,11
1655 | uH5eTFU6eRM_00092.jpg,11
1656 | 9ldPdyxkca4_04658.jpg,119
1657 | _hau617Vex8_05132.jpg,11
1658 | T3sgbipBWwg_12477.jpg,11
1659 | cosWi0n6kgk_00988.jpg,11
1660 | rYKtWL0bkkg_01737.jpg,11
1661 | SUVYuq1kNOI_09102.jpg,11
1662 | GDK7ly_JwYc_04207.jpg,8
1663 | WLexzfEsc2Y_10571.jpg,11
1664 | oTVKckgd7D0_02815.jpg,8
1665 | _XSRRMW5TC4_03399.jpg,11
1666 | rtwQPFg4YBQ_00606.jpg,11
1667 | eMQ1G4hTS1w_10614.jpg,11
1668 | EwStEi2ScG8_02824.jpg,25
1669 | C8MgH9u5Q6o_01038.jpg,11
1670 | qvOZV2x-ErQ_01276.jpg,128
1671 | 2vwyMZDYsx0_04185.jpg,94
1672 | PDaW6kohY0o_06468.jpg,7
1673 | eygquDTqRHo_02470.jpg,11
1674 | 2uFU5KFsUhs_01470.jpg,11
1675 | 6xftdgXU3DA_00343.jpg,11
1676 | QaLmTRpFe3Y_12986.jpg,44
1677 | _hau617Vex8_00518.jpg,11
1678 | 71bCN2RvX7w_01205.jpg,11
1679 | b-w6yVn-fFc_01044.jpg,24
1680 | 3eylU7OeIBk_02067.jpg,11
1681 | 2uFU5KFsUhs_01826.jpg,11
1682 | Jd3A1Oh3KgQ_01353.jpg,7
1683 | 4VEy6RP45Pk_03054.jpg,44
1684 | HfNlRgtxoT0_07924.jpg,11
1685 | 2tWkHiX8K5o_01700.jpg,11
1686 | BW1IsHNn8iM_08435.jpg,11
1687 | r7Gwaoru_gg_01397.jpg,6
1688 | _TugE6xRTEY_03013.jpg,11
1689 | ewkz036a944_02261.jpg,11
1690 | tTOkndNGbso_02063.jpg,11
1691 | Kz49kdjmNFA_01733.jpg,11
1692 | S_HrZ5r1EzQ_03202.jpg,204
1693 | fuokiCZ0FOA_02557.jpg,135
1694 | 9TJVA2lwW4A_03014.jpg,25
1695 | 6jXN80cNWNs_09586.jpg,7
1696 | d1tV9SxrccY_01036.jpg,82
1697 | 1eTSE4U2QK0_00976.jpg,11
1698 | oM2QJ7ABrEI_01506.jpg,11
1699 | AbZUxRsE6EM_01079.jpg,44
1700 | kD_yUKB85HM_00000.jpg,11
1701 | _WzBGnioCls_00603.jpg,25
1702 | 2uucpO_3f2k_06638.jpg,24
1703 | 33T7MltfPGE_01028.jpg,28
1704 | VIS-okyTpIM_08190.jpg,113
1705 | hKWosn-YZ3o_04240.jpg,11
1706 | eMQ1G4hTS1w_02114.jpg,11
1707 | F4ovVbk4hP0_03501.jpg,7
1708 | _hau617Vex8_00576.jpg,11
1709 | 9ERpBDev2oM_00000.jpg,135
1710 | 65mQtjchFdQ_00150.jpg,114
1711 | 5IuRzJRrRpQ_00955.jpg,11
1712 | 0Z65qwDms7U_00233.jpg,8
1713 | TDg3pEah-Gc_04591.jpg,11
1714 | 2tWkHiX8K5o_00190.jpg,135
1715 | DHj9j1UYbAA_09762.jpg,42
1716 | r7Gwaoru_gg_02313.jpg,128
1717 | 5ySqNOX6KxE_02198.jpg,52
1718 | qnq7euJ73kw_08951.jpg,11
1719 | b-MbEx1Iqj8_00270.jpg,128
1720 | ufwz8nDkI_Q_02818.jpg,19
1721 | 2uucpO_3f2k_06553.jpg,11
1722 | 6_sLF3OBtc8_02513.jpg,11
1723 | 778rYLWbcMs_01142.jpg,11
1724 | 3V56VXL0PPs_09712.jpg,11
1725 | cYGytFPWcjE_04210.jpg,11
1726 | ijIxKUAgB7w_02332.jpg,94
1727 | Abz4ZbcESKA_01636.jpg,11
1728 | F4ovVbk4hP0_01604.jpg,7
1729 | 6Aj8v2QmjcA_09421.jpg,11
1730 | KFCodkH917k_07147.jpg,8
1731 | 7I_X57GcSTw_04040.jpg,135
1732 | LVkNEkr0QPQ_01316.jpg,11
1733 | 3zCAohJUH50_01792.jpg,6
1734 | ojhKtzwJTm4_06913.jpg,11
1735 | ewkz036a944_01777.jpg,24
1736 | 58fWNmsEM1A_01672.jpg,7
1737 | Abz4ZbcESKA_01815.jpg,11
1738 | xLsFUtbX51Y_04615.jpg,8
1739 | QMnn4l0FwG4_01725.jpg,7
1740 | So5tgIbW81o_04889.jpg,156
1741 | D9CaGSKGzbk_00470.jpg,44
1742 | tCNgetoOFjE_02992.jpg,8
1743 | xLsFUtbX51Y_04376.jpg,97
1744 | tCNgetoOFjE_03022.jpg,8
1745 | Du9eKNTgAFg_04102.jpg,8
1746 | D9CaGSKGzbk_02454.jpg,8
1747 | A60F1488bxw_01723.jpg,7
1748 | 455oSK3nA2I_01021.jpg,7
1749 | ojhKtzwJTm4_06262.jpg,8
1750 | yzRZDzLTwrc_14160.jpg,7
1751 | r0CwydJ3_P4_05560.jpg,8
1752 | oVrKzYP476s_01555.jpg,8
1753 | mZgSZBTTRcA_00060.jpg,8
1754 | 1P9VOVXsoLA_10635.jpg,8
1755 | oWt8IXACnqg_01288.jpg,8
1756 | 4zqYCG7Tg70_02453.jpg,8
1757 | A1f1kvKAsKM_03905.jpg,7
1758 | r0CwydJ3_P4_11221.jpg,8
1759 | So5tgIbW81o_05040.jpg,8
1760 | htBRpB64sYQ_14009.jpg,144
1761 | 6cIi8rdzj50_03935.jpg,8
1762 | r_Gk1whxRs8_02338.jpg,97
1763 | 8z8cI8EFt-g_02258.jpg,8
1764 | skYLmxFviG8_06762.jpg,8
1765 | uJXsiyov2AM_03818.jpg,8
1766 | 1P9VOVXsoLA_14414.jpg,8
1767 | tCALjANX8B4_05300.jpg,8
1768 | 4zqYCG7Tg70_01573.jpg,8
1769 | 37LHdSJOKmQ_00671.jpg,171
1770 | yzRZDzLTwrc_09604.jpg,8
1771 | skYLmxFviG8_03177.jpg,8
1772 | v0hjtKzxvRU_05921.jpg,7
1773 | RnAh_yMaVdE_00091.jpg,8
1774 | sYqzl4xH4Ts_10444.jpg,8
1775 | CW22ZLfuY_0_02363.jpg,8
1776 | iDlLCKnOB2M_02325.jpg,97
1777 | iDlLCKnOB2M_09658.jpg,8
1778 | amwE4tnOUVA_00100.jpg,8
1779 | 6cIi8rdzj50_03423.jpg,8
1780 | shGA-qurh_E_00641.jpg,8
1781 | 8z8cI8EFt-g_06028.jpg,8
1782 | zYGjzajwNIs_06435.jpg,8
1783 | r0CwydJ3_P4_02572.jpg,157
1784 | r0CwydJ3_P4_08369.jpg,8
1785 | IZhSi0WJtKM_04285.jpg,8
1786 | nMIfsnmNGK8_00279.jpg,7
1787 | 1P9VOVXsoLA_03975.jpg,8
1788 | ewr-vzw4ZWo_04083.jpg,8
1789 | oWt8IXACnqg_02564.jpg,8
1790 | tCALjANX8B4_01047.jpg,8
1791 | k5R_XwLso0Q_05397.jpg,8
1792 | 9c2Atsc4JnM_06287.jpg,97
1793 | JISPuZok08I_02950.jpg,8
1794 | yvLvoi3h6lU_15764.jpg,8
1795 | JISPuZok08I_02537.jpg,8
1796 | _XSRRMW5TC4_01843.jpg,8
1797 | IxLQ2vjgNZA_03407.jpg,8
1798 | DrpNgJU3lwI_02185.jpg,97
1799 | s2SPB1xdxBY_11709.jpg,8
1800 | 6cIi8rdzj50_02741.jpg,8
1801 | _XSRRMW5TC4_03124.jpg,8
1802 | tCALjANX8B4_03126.jpg,50
1803 | CW22ZLfuY_0_03137.jpg,7
1804 | rSznvuywzEY_03006.jpg,8
1805 | JISPuZok08I_02785.jpg,8
1806 | _duOR6K5xbw_08620.jpg,19
1807 | Q1N-RTIR1dk_02235.jpg,8
1808 | uJXsiyov2AM_03161.jpg,8
1809 | 0NIoPNiUt7o_03896.jpg,8
1810 | 8z8cI8EFt-g_01267.jpg,8
1811 | 1P9VOVXsoLA_01505.jpg,8
1812 | DrpNgJU3lwI_03729.jpg,97
1813 | G-XP4k7jFKc_03369.jpg,8
1814 | oTVKckgd7D0_00181.jpg,8
1815 | evBGqAiuB7E_02556.jpg,8
1816 | A-faE9Gu2c0_03761.jpg,114
1817 | k5R_XwLso0Q_07502.jpg,8
1818 | rSznvuywzEY_07842.jpg,8
1819 | DjPv6uKLSPY_02307.jpg,8
1820 | So5tgIbW81o_04563.jpg,8
1821 | oTVKckgd7D0_09010.jpg,8
1822 | A60F1488bxw_01822.jpg,8
1823 | RnAh_yMaVdE_01473.jpg,8
1824 | skYLmxFviG8_06784.jpg,8
1825 | UdYfTV0sCSM_02809.jpg,8
1826 | 5BZh3OCuoS0_02861.jpg,8
1827 | kSHo9xhrpdk_00236.jpg,8
1828 | zYGjzajwNIs_06481.jpg,8
1829 | ewr-vzw4ZWo_04520.jpg,8
1830 | G-XP4k7jFKc_03273.jpg,8
1831 | K62-WvsPUjg_02401.jpg,8
1832 | YLPVXcgbNSQ_00890.jpg,8
1833 | rmcQSeAa4M0_02697.jpg,8
1834 | 9w4Rapjz-NM_03285.jpg,8
1835 | 4zqYCG7Tg70_01549.jpg,8
1836 | r0CwydJ3_P4_13463.jpg,8
1837 | G-XP4k7jFKc_03240.jpg,8
1838 | tCNgetoOFjE_04651.jpg,6
1839 | qkNSogHOg9M_01726.jpg,42
1840 | 3BfNRUOoNIE_03193.jpg,42
1841 | qP0wcVHLU4c_00662.jpg,42
1842 | D8AQfyoMjpk_04418.jpg,42
1843 | cOv2TDLv8wU_10329.jpg,42
1844 | D8AQfyoMjpk_01758.jpg,42
1845 | cOv2TDLv8wU_04650.jpg,42
1846 | qP0wcVHLU4c_02961.jpg,42
1847 | jQKQOuJt7SE_07072.jpg,42
1848 | E8IwAjMHrh4_02363.jpg,42
1849 | 5mjJLh5oqSw_00653.jpg,42
1850 | s0N81VG1R_g_01554.jpg,42
1851 | QuRWCGH21Wk_03584.jpg,42
1852 | _Ynt54lyJGw_00331.jpg,42
1853 | o5w6QUCPm2w_00000.jpg,42
1854 | qkNSogHOg9M_01800.jpg,42
1855 | Mo3zOXdWEtk_01346.jpg,42
1856 | zvSn1nIDXfg_02333.jpg,42
1857 | CAWE87eRD6c_00116.jpg,0
1858 | 2Tzk6dDTs9c_01077.jpg,42
1859 | QuRWCGH21Wk_03298.jpg,42
1860 | e-aFP_PxOd8_00000.jpg,76
1861 | 2Tzk6dDTs9c_01294.jpg,42
1862 | lyIKMg__Yss_00010.jpg,42
1863 | _Ynt54lyJGw_01165.jpg,42
1864 | wBDk55xgPSA_00204.jpg,42
1865 | THERX5WLPCI_00432.jpg,42
1866 | 6-SKQf6EN_s_02461.jpg,42
1867 | YBM99Z1a2jc_06201.jpg,42
1868 | wBDk55xgPSA_00000.jpg,2
1869 | 6-SKQf6EN_s_03758.jpg,0
1870 | AY4tGuV7xhc_00348.jpg,42
1871 | 6-SKQf6EN_s_02623.jpg,42
1872 | pTH8QHmpoNA_01000.jpg,42
1873 | SryV1cz_ZMA_03729.jpg,42
1874 | U94qRFmHoo4_04161.jpg,42
1875 | MqzBgu91wBs_02210.jpg,42
1876 | 6-SKQf6EN_s_07759.jpg,42
1877 | grZz7WzbBXM_02000.jpg,42
1878 | Gy-bPvUMZos_00000.jpg,42
1879 | 6-SKQf6EN_s_05222.jpg,42
1880 | cOv2TDLv8wU_08241.jpg,42
1881 | oVb_diFDfmE_00928.jpg,42
1882 | cOv2TDLv8wU_01406.jpg,42
1883 | ivXmXvYSoNc_00000.jpg,42
1884 | REbQ_07yLlU_02708.jpg,42
1885 | 2xjkyIZSjvM_04000.jpg,42
1886 | _Ynt54lyJGw_01178.jpg,42
1887 | 7t8BzhTqwB0_04113.jpg,42
1888 | 6-SKQf6EN_s_00369.jpg,42
1889 | KOuWAMukPjM_03139.jpg,42
1890 | UJsN7VGqHBo_03336.jpg,57
1891 | 7t8BzhTqwB0_05122.jpg,42
1892 | 6-SKQf6EN_s_06129.jpg,42
1893 | CJF4dxfDIkc_05462.jpg,42
1894 | Z4tSR8IzC2o_00000.jpg,42
1895 | s0N81VG1R_g_02666.jpg,42
1896 | MqzBgu91wBs_02921.jpg,42
1897 | czYNJkSzl-M_00388.jpg,52
1898 | wxjWmvn9vjs_00346.jpg,57
1899 | aLTIfqknNRU_04459.jpg,42
1900 | 7ljaPOLO-Tk_01060.jpg,42
1901 | WWqZlnMz9yE_00000.jpg,42
1902 | 6-SKQf6EN_s_06400.jpg,42
1903 | D8AQfyoMjpk_01459.jpg,42
1904 | KOuWAMukPjM_08572.jpg,42
1905 | qkNSogHOg9M_01696.jpg,76
1906 | bH8oqs0vKxg_00174.jpg,52
1907 | 6-SKQf6EN_s_07445.jpg,42
1908 | SryV1cz_ZMA_01270.jpg,42
1909 | afZ4bpFm7Tg_01000.jpg,42
1910 | qkNSogHOg9M_02684.jpg,42
1911 | p5xH7ovUqCE_02076.jpg,42
1912 | 6-SKQf6EN_s_07576.jpg,0
1913 | eohfjLMHDmg_02716.jpg,42
1914 | 6-SKQf6EN_s_07534.jpg,42
1915 | YBM99Z1a2jc_00824.jpg,42
1916 | E5ThTSYt-rQ_06724.jpg,42
1917 | E8IwAjMHrh4_01821.jpg,42
1918 | KOuWAMukPjM_07242.jpg,42
1919 | rAxS3LEe9Xs_00153.jpg,0
1920 | L-YM6VzZ_d4_03387.jpg,42
1921 | cOv2TDLv8wU_02518.jpg,42
1922 | 6-SKQf6EN_s_00459.jpg,42
1923 | 6-SKQf6EN_s_00298.jpg,42
1924 | E8IwAjMHrh4_01475.jpg,42
1925 | qaFKf9wJY2U_00553.jpg,42
1926 | 5FrNjG0xmSs_00261.jpg,42
1927 | qP0wcVHLU4c_09342.jpg,42
1928 | 6-SKQf6EN_s_08840.jpg,42
1929 | 0O200AVGGko_01583.jpg,42
1930 | 6QA_tCJFa3Q_00203.jpg,42
1931 | YBM99Z1a2jc_05437.jpg,42
1932 | NZNc12T3N8E_01551.jpg,6
1933 | CDFbfUnr2lE_07999.jpg,171
1934 | PjIkgSK2bbM_10000.jpg,98
1935 | VuHzy670qQw_05794.jpg,171
1936 | YjqMAmOdPJE_10903.jpg,98
1937 | Uiz4qNneH-4_12936.jpg,8
1938 | Kpq3UXNXgHI_09025.jpg,171
1939 | PjIkgSK2bbM_01082.jpg,98
1940 | MRIkMhtf53A_02692.jpg,1
1941 | m3Dmgxa5Kzo_06126.jpg,98
1942 | MRIkMhtf53A_05091.jpg,6
1943 | 15P2Pisr7K8_01688.jpg,98
1944 | LFZFbrcrs68_03487.jpg,98
1945 | skGJwu9trsI_04398.jpg,171
1946 | 15P2Pisr7K8_06596.jpg,1
1947 | 5Zp5SDOAp0c_11115.jpg,98
1948 | cRqZAhoDNeE_14623.jpg,98
1949 | fcyMJyBfDUo_01615.jpg,127
1950 | oVp5taUCAlc_01784.jpg,98
1951 | oXCR7twx3O8_03727.jpg,98
1952 | j1q79eA3wR4_00581.jpg,98
1953 | CcSHovb-BMQ_00717.jpg,98
1954 | Uiz4qNneH-4_09025.jpg,171
1955 | itmWPv2L4-U_07779.jpg,98
1956 | WKlOhAjyr2A_02576.jpg,171
1957 | oeFSIb_hM8o_07737.jpg,19
1958 | 6KRJouIDEaU_00501.jpg,19
1959 | bk4aFTqS0Rk_00565.jpg,19
1960 | dQSdI0s1bg4_01264.jpg,19
1961 | El8A38OXcLM_01821.jpg,19
1962 | FeJfdKb52E0_02262.jpg,19
1963 | DMKm0r7eZXY_00897.jpg,19
1964 | xTSBhaYHqJw_08618.jpg,19
1965 | Em6n0n4t1kI_09215.jpg,19
1966 | 8OFBbsVAeXQ_03000.jpg,19
1967 | XM6q4Qj-5bw_06624.jpg,89
1968 | dQSdI0s1bg4_01365.jpg,19
1969 | v9OUTZQuZAU_01632.jpg,19
1970 | xTSBhaYHqJw_00047.jpg,19
1971 | AT7p0xgZlG0_04339.jpg,19
1972 | xTSBhaYHqJw_00672.jpg,19
1973 | esvbJvODlUQ_05239.jpg,19
1974 | oeFSIb_hM8o_08119.jpg,19
1975 | 0VfWNUlf1H0_01418.jpg,19
1976 | Em6n0n4t1kI_08928.jpg,19
1977 | 8YckoGvJCSg_00317.jpg,19
1978 | jUavI4IJ60U_03586.jpg,8
1979 | ofBXXPGooCk_04340.jpg,19
1980 | 5RgS1sEBK30_00793.jpg,19
1981 | 0vmwXPoxtcQ_01425.jpg,19
1982 | qLVdu5olfn0_03238.jpg,19
1983 | htp69IS3KAE_04984.jpg,19
1984 | ofBXXPGooCk_02633.jpg,19
1985 | esvbJvODlUQ_02402.jpg,19
1986 | WOl6b5yMn5U_02764.jpg,19
1987 | esvbJvODlUQ_00811.jpg,19
1988 | Em6n0n4t1kI_06734.jpg,19
1989 | nxYHR-z4RII_03975.jpg,24
1990 | Wd1tDcplTic_02466.jpg,6
1991 | 1kO0iAeKWuM_06097.jpg,52
1992 | jK0j4tiyytw_01108.jpg,52
1993 | uKVnjnTSv_A_08190.jpg,52
1994 | Z-DP85lP0Mg_05312.jpg,52
1995 | 5ppwCRcnvjs_02109.jpg,52
1996 | fYsuBmk2ciY_10796.jpg,128
1997 | B_UwTeUZ5tU_00218.jpg,128
1998 | W0V-hB-RGdM_01753.jpg,52
1999 | H-cVpDZbHV0_01563.jpg,113
2000 | _EfDVO9ca90_03073.jpg,52
2001 | iMYqAKwdHJQ_07273.jpg,52
2002 | QYx1IrMutPI_02449.jpg,128
2003 | hCRJCCmSgCU_09142.jpg,52
2004 | 8WY49-YvU00_00705.jpg,113
2005 | 7EOuX8dexqw_07707.jpg,52
2006 | TXkPCy8KMTc_01538.jpg,52
2007 | mrnoeZsBx58_04007.jpg,52
2008 | ffPIhZs4l4I_01539.jpg,128
2009 | V8xsZ50-93E_01588.jpg,24
2010 | NIKCstu89zk_04250.jpg,52
2011 | BAkqdr-H0tw_09916.jpg,52
2012 | 9PVD3wcACeU_03589.jpg,128
2013 | NADszG9W2K4_01984.jpg,52
2014 | 2irSP-6ftho_00230.jpg,52
2015 | xIS5_lw0jIQ_00494.jpg,128
2016 | 69Ap8WsenXc_05189.jpg,52
2017 | Eiznagnni28_06448.jpg,24
2018 | Z-DP85lP0Mg_04752.jpg,52
2019 | 26hhv1v9b1U_03980.jpg,158
2020 | BBJNmhfhQRw_03397.jpg,52
2021 | bo1KQQv6XTA_00377.jpg,52
2022 | nVyxLqHrAfA_06644.jpg,52
2023 | ZJRfMdu5jAk_01805.jpg,52
2024 | 6UIae8IsfSI_04511.jpg,52
2025 | o6jpD1dqB7Q_03004.jpg,52
2026 | iTq9CxK6gZc_04424.jpg,52
2027 | 6bHlRlaW3T8_01874.jpg,52
2028 | Va4qK_XbvCc_02997.jpg,52
2029 | gp01NXa0vS4_04549.jpg,52
2030 | HpaqpfduVAs_00215.jpg,128
2031 | gp01NXa0vS4_07006.jpg,52
2032 | lfJ1QhSu4Rw_07326.jpg,0
2033 | YD0hXVEu-NQ_00200.jpg,113
2034 | 4Eqfad9F47g_02088.jpg,52
2035 | Mq7WfGHHZOc_08297.jpg,52
2036 | 4Eqfad9F47g_01071.jpg,52
2037 | xIS5_lw0jIQ_05180.jpg,128
2038 | T9aJiu8yur8_00701.jpg,128
2039 | YdDnMQ3pLz4_01348.jpg,128
2040 | SFZ39w4lx0c_01454.jpg,52
2041 | XBbvF6lgTD4_02083.jpg,52
2042 | j89BWbH4WHY_05217.jpg,52
2043 | 26hhv1v9b1U_04611.jpg,52
2044 | dML3pRVq3E8_05874.jpg,128
2045 | 1MZOTl3AJ0s_13353.jpg,113
2046 | fYsuBmk2ciY_11195.jpg,52
2047 | 1XN00Gvpdv8_03973.jpg,52
2048 | xfk6H1kHtec_00874.jpg,15
2049 | Wt9V_NZqHBk_05609.jpg,52
2050 | Z-DP85lP0Mg_01900.jpg,128
2051 | gp01NXa0vS4_06840.jpg,113
2052 | XxiEESGGJgc_01884.jpg,15
2053 | QctPDZujLSg_00644.jpg,15
2054 | 9m4i1zJDPdU_04386.jpg,15
2055 | eFvpFp0SRiw_05445.jpg,15
2056 | C2aWqmIJlds_00174.jpg,15
2057 | 7AjmaDsauVY_00735.jpg,15
2058 | z6MBCh7o2kA_01455.jpg,15
2059 | rzhw6a5CBWA_04605.jpg,15
2060 | svfQtbhIdD8_00210.jpg,15
2061 | cUH18dDr-Kw_05919.jpg,89
2062 | C2aWqmIJlds_00616.jpg,20
2063 | 5hKd22FpHvU_00933.jpg,15
2064 | BKt9a8LNdSE_05408.jpg,15
2065 | DGrDF0ScE5I_05430.jpg,15
2066 | u2iO55GnsCs_07685.jpg,15
2067 | DGrDF0ScE5I_03833.jpg,15
2068 | cLZJMBTzRnE_01139.jpg,15
2069 | B9G88j18zDw_02218.jpg,165
2070 | b6YGmXBYj1c_04772.jpg,165
2071 | Qwhtg63OTSM_00180.jpg,165
2072 | b6YGmXBYj1c_07044.jpg,165
2073 | glQcUkV48_s_15245.jpg,165
2074 | AqTNE15CpX4_01010.jpg,190
2075 | AqTNE15CpX4_05366.jpg,165
2076 | pofXacIBqrc_00953.jpg,165
2077 | pofXacIBqrc_01068.jpg,165
2078 | onv5GDNJIhY_06144.jpg,165
2079 | K-3bCj2EQDM_02361.jpg,165
2080 | AqTNE15CpX4_04692.jpg,165
2081 | b6YGmXBYj1c_07641.jpg,165
2082 | B9G88j18zDw_03225.jpg,165
2083 | Du9eKNTgAFg_03108.jpg,165
2084 | onv5GDNJIhY_06202.jpg,165
2085 | onv5GDNJIhY_06266.jpg,165
2086 | B9G88j18zDw_03997.jpg,165
2087 | P0KYga1YQo0_02747.jpg,165
2088 | gtnuQCniXS0_02562.jpg,165
2089 | gtnuQCniXS0_02405.jpg,165
2090 | b6YGmXBYj1c_07295.jpg,165
2091 | q3_e-BxoJX8_01854.jpg,165
2092 | b6YGmXBYj1c_07126.jpg,165
2093 | x_L_ykc6mGw_09641.jpg,165
2094 | b6YGmXBYj1c_07596.jpg,165
2095 | AqTNE15CpX4_01922.jpg,6
2096 | BNV91lUuqms_01762.jpg,165
2097 | b6YGmXBYj1c_07492.jpg,165
2098 | KIOq4hQqxsc_15090.jpg,165
2099 | onv5GDNJIhY_06081.jpg,165
2100 | gtnuQCniXS0_03620.jpg,165
2101 | pofXacIBqrc_00607.jpg,165
2102 | onv5GDNJIhY_06647.jpg,165
2103 | dbMFEIku4dI_00269.jpg,24
2104 | 4EZOodrNWgU_05944.jpg,24
2105 | xb7AvTWMJ64_01528.jpg,24
2106 | fBFNaCGu4Ks_01503.jpg,113
2107 | 1FRZJlnzOaw_03563.jpg,24
2108 | 6qKAj_IMUGY_01859.jpg,24
2109 | g8906kr_V8Y_01316.jpg,24
2110 | 8R-uTuekyq0_02501.jpg,24
2111 | YITVOA7MDnI_01614.jpg,24
2112 | 6bsQuifkRSU_06197.jpg,24
2113 | XGKYlPITn6U_00932.jpg,127
2114 | 2sXXdDcK-AE_02233.jpg,24
2115 | 6Xc_RHAw3ds_00713.jpg,24
2116 | eHXuEazoYyQ_00631.jpg,24
2117 | N0lMyIvxNTs_04396.jpg,7
2118 | Y4EASeZ6IeM_01370.jpg,52
2119 | yuQfELoZ9fo_02502.jpg,24
2120 | h95H-ZV42EM_01625.jpg,24
2121 | eHXuEazoYyQ_01200.jpg,114
2122 | 9UyOprSFAro_01744.jpg,11
2123 | EHWfonQfz1Y_01662.jpg,24
2124 | Xrtplc_OE9I_00925.jpg,24
2125 | dbMFEIku4dI_00384.jpg,24
2126 | wyR64tVPsKg_00504.jpg,24
2127 | 6qKAj_IMUGY_01765.jpg,24
2128 | wSpLHsGk4so_06637.jpg,24
2129 | Tte4w3ob_Fo_00873.jpg,24
2130 | k7hFiK8Rh-4_14653.jpg,24
2131 | ypSBAGUH9xk_00325.jpg,24
2132 | e9o6IrLQSrI_01237.jpg,52
2133 | iFS9RsWwKSo_00498.jpg,24
2134 | YITVOA7MDnI_01625.jpg,24
2135 | qF_gvlAl0-Q_00309.jpg,128
2136 | vxfPyvSdCK8_00191.jpg,24
2137 | Yau8EWeq5PQ_00217.jpg,24
2138 | 9J_pQ5kCpK8_00900.jpg,24
2139 | hAdImP_rAj4_00276.jpg,24
2140 | qG682w9sqwY_00148.jpg,24
2141 | 3Yd0uQ3rnos_04357.jpg,24
2142 | 3Zz4Z80iK7c_01080.jpg,24
2143 | wyR64tVPsKg_00242.jpg,24
2144 | 1P6lirTg6Ds_00723.jpg,24
2145 | qVO89xE8o1I_01043.jpg,24
2146 | qCsMHJMgkYo_00266.jpg,24
2147 | wJxmOwN7eS4_04389.jpg,24
2148 | o3T7paQTcjc_01038.jpg,24
2149 | oZP1AcbJozM_01073.jpg,24
2150 | wGAAzulconQ_10087.jpg,24
2151 | hLtRy8HCWfY_00587.jpg,24
2152 | jwm4UXcxf2U_00767.jpg,24
2153 | YITVOA7MDnI_00995.jpg,24
2154 | 6pn7efepRag_03521.jpg,24
2155 | n00qAOF1grI_02414.jpg,24
2156 | TS9ixewwZqA_05922.jpg,24
2157 | p5AjcMyOQi8_02211.jpg,24
2158 | zeMmB4X_T8c_02646.jpg,128
2159 | 1EV2tLKJpyU_07898.jpg,24
2160 | PUd6KlCGHk8_01803.jpg,24
2161 | uDocbJe12OQ_00749.jpg,24
2162 | 5u_k1rWk32A_03863.jpg,24
2163 | _f1wF169Ffk_07463.jpg,24
2164 | BCzmcS1jrbs_01310.jpg,52
2165 | 2sXXdDcK-AE_02761.jpg,24
2166 | h95H-ZV42EM_00647.jpg,127
2167 | ZzwpoO7R3eQ_02927.jpg,24
2168 | 4EZOodrNWgU_02942.jpg,24
2169 | ythmWU4Nen0_00950.jpg,24
2170 | 9UyOprSFAro_04889.jpg,11
2171 | qx3lKqRexhU_01386.jpg,24
2172 | uzpWFybfLRc_15606.jpg,24
2173 | 4EZOodrNWgU_04717.jpg,127
2174 | wGAAzulconQ_01602.jpg,24
2175 | SmtKPPhWRWI_01299.jpg,11
2176 | eHXuEazoYyQ_00045.jpg,24
2177 | G3OZvsFdoF0_09125.jpg,24
2178 | OccwUfG2114_05270.jpg,24
2179 | 6bsQuifkRSU_12349.jpg,52
2180 | rD5xqo3Xbu0_00332.jpg,24
2181 | ythmWU4Nen0_02313.jpg,24
2182 | n00qAOF1grI_09555.jpg,24
2183 | rXCfcQXv54I_01748.jpg,127
2184 | aEUbl9oT_0s_04257.jpg,24
2185 | s1bFpZ_Y7Hg_00000.jpg,24
2186 | sLOPgfRyxHg_04394.jpg,24
2187 | 6t_tv8CAKqY_02222.jpg,24
2188 | hlpw1CpBleM_08033.jpg,24
2189 | vUm8xfi9avY_05093.jpg,6
2190 | 5iR3TYgTzxw_05056.jpg,114
2191 | X4y8TCAkOZ0_01477.jpg,6
2192 | GHc0nzcuwKg_02552.jpg,6
2193 | 3p5pkqoqHlg_10001.jpg,7
2194 | _89ktPIOv2E_02935.jpg,6
2195 | LjPZjjaQTq4_10797.jpg,6
2196 | 5U82Q8shYwI_09843.jpg,6
2197 | SobjxWRibE0_00909.jpg,6
2198 | HAvRDzAc_Ig_02282.jpg,6
2199 | GHc0nzcuwKg_02513.jpg,114
2200 | gzJtsi7OKtQ_06694.jpg,6
2201 | GYldJ3fRkjs_04466.jpg,6
2202 | VNfCm5s7U9Y_07103.jpg,7
2203 | HkcThBTNAxY_03808.jpg,6
2204 | vUm8xfi9avY_01464.jpg,6
2205 | wq-L-ELl8Vk_00477.jpg,7
2206 | AiTCgjr08so_06224.jpg,6
2207 | 1DniGYghWmU_05024.jpg,6
2208 | HAvRDzAc_Ig_03930.jpg,6
2209 | wz4CodOZVpM_08006.jpg,6
2210 | wz4CodOZVpM_02615.jpg,6
2211 | ftkDx8emU80_05606.jpg,6
2212 | datvszwva68_03756.jpg,6
2213 | 5U82Q8shYwI_09968.jpg,6
2214 | X2RjucWZ2ik_05855.jpg,6
2215 | MUdNPozh1VA_02885.jpg,6
2216 | 9zJEVBpsNe0_00025.jpg,5
2217 | SntkmfYUoW0_05237.jpg,6
2218 | SmIcu_tigtU_05037.jpg,6
2219 | datvszwva68_00053.jpg,6
2220 | _kjA0fcEiUU_04937.jpg,7
2221 | v53aJHx6Ajk_01768.jpg,6
2222 | yGB_7MHkcIU_02672.jpg,98
2223 | U7IhX-lf1Ak_00255.jpg,6
2224 | Th8KYJNyxeY_11022.jpg,6
2225 | eQA0sfMhNis_04062.jpg,6
2226 | MUdNPozh1VA_00771.jpg,6
2227 | uYdY4K1YUSY_06669.jpg,8
2228 | yXnF9SiU1AU_00623.jpg,6
2229 | QsvOxoUd1_A_05175.jpg,6
2230 | qwuzZkGbhBQ_00841.jpg,6
2231 | dGSETw8NjZ4_00789.jpg,6
2232 | yjnaVrrPZqk_04233.jpg,6
2233 | yjGi-6tSzGs_04259.jpg,50
2234 | LjPZjjaQTq4_01527.jpg,6
2235 | 7tZx0jUHc8c_05965.jpg,6
2236 | X2RjucWZ2ik_13685.jpg,6
2237 | B4adkjC_ZBU_06314.jpg,6
2238 | HLB23U3Leec_11322.jpg,6
2239 | r4k0eZslNV8_00103.jpg,6
2240 | B4adkjC_ZBU_10060.jpg,6
2241 | lTAlBoWIhAo_03264.jpg,6
2242 | DK1wUzGHdWI_05895.jpg,6
2243 | 5U82Q8shYwI_10248.jpg,6
2244 | ssjYnPCnXGs_01441.jpg,6
2245 | MYh02-UmnGs_06159.jpg,7
2246 | HLB23U3Leec_04764.jpg,8
2247 | SmIcu_tigtU_03420.jpg,6
2248 | Ph_Be9ZSeR0_03459.jpg,6
2249 | gHVZOwYmb0g_00787.jpg,6
2250 | zsN6jS4Wy4w_05699.jpg,6
2251 | Lay-eCE2OdI_00578.jpg,6
2252 | C4J0L7i7fGM_01400.jpg,6
2253 | JWMEDIyY710_02048.jpg,6
2254 | yLUxzgdeD78_11475.jpg,6
2255 | bCDqgnrspRQ_00319.jpg,6
2256 | xUbfV_ZXSXk_04610.jpg,135
2257 | ifPH7hEztUE_07478.jpg,6
2258 | lvveZFnPAcY_02475.jpg,6
2259 | JWMEDIyY710_07727.jpg,6
2260 | LjPZjjaQTq4_05823.jpg,6
2261 | SmIcu_tigtU_05294.jpg,6
2262 | GOsL1WHJI4w_00691.jpg,6
2263 | yXnF9SiU1AU_00378.jpg,6
2264 | X2RjucWZ2ik_07480.jpg,6
2265 | 8F4kHJ3f3xE_05288.jpg,6
2266 | mnbiRWI_lcE_10589.jpg,98
2267 | cs7-Tw7eocI_00425.jpg,6
2268 | Z8mJWia5omE_01057.jpg,8
2269 | Yb82-solv5U_02252.jpg,98
2270 | _l8vVkRpSdY_03448.jpg,6
2271 | Ba39k5Zrias_03303.jpg,23
2272 | bVwzdRWjt9U_08997.jpg,23
2273 | 4lFkbJphYm8_01443.jpg,23
2274 | WYVKLOi5p5Q_08081.jpg,113
2275 | 3RHWcxfxB6o_02108.jpg,23
2276 | b59REMZXc3c_03860.jpg,23
2277 | DdDW5lelkRI_09010.jpg,23
2278 | AQW_jG3W7WI_12013.jpg,23
2279 | ZHSPLH5-zvw_05082.jpg,23
2280 | YidNRS2_xqM_06762.jpg,23
2281 | 5bk-knXKYmY_07183.jpg,23
2282 | Ba39k5Zrias_02824.jpg,23
2283 | ZHSPLH5-zvw_03904.jpg,23
2284 | H8A02mx3dAI_04694.jpg,23
2285 | MEnHPTbf6dE_02107.jpg,23
2286 | ZS4PfHJeFyc_02306.jpg,23
2287 | GS89imYSZR8_03101.jpg,23
2288 | MEnHPTbf6dE_01635.jpg,23
2289 | ExAe4W3CdGY_02479.jpg,23
2290 | 0nGOeKwpGH4_00842.jpg,23
2291 | Yjz_fuO8zVA_05370.jpg,23
2292 | LzgrNllEqPA_03537.jpg,23
2293 | 6ZY8raPkDys_06802.jpg,23
2294 | IMFfJ6dZb2g_03673.jpg,23
2295 | XY9h2Ot1zR8_01684.jpg,0
2296 | LQ6pXqkv-5s_05839.jpg,23
2297 | dfDP_2MEfI4_03930.jpg,23
2298 | BbOZCMbryI0_01027.jpg,23
2299 | _r0syG5xAlg_05346.jpg,23
2300 | f_cOlUKIYy0_02764.jpg,23
2301 | ThTh0ptLwfo_00261.jpg,82
2302 | Kve7o5wcev4_04112.jpg,23
2303 | U-yDIOh6AmU_04056.jpg,23
2304 | 5bk-knXKYmY_06586.jpg,23
2305 | ZS4PfHJeFyc_05375.jpg,23
2306 | VWX7iwVc0Oc_01008.jpg,23
2307 | CM5pNHmoNEo_04446.jpg,23
2308 | CM5pNHmoNEo_00113.jpg,23
2309 | 5bk-knXKYmY_03306.jpg,23
2310 | __kupr7KQos_05262.jpg,23
2311 | GS89imYSZR8_04231.jpg,23
2312 | YidNRS2_xqM_04513.jpg,23
2313 | _3Lb-lGfkL4_05217.jpg,23
2314 | TuAGqlnt7B8_14984.jpg,23
2315 | GS89imYSZR8_04268.jpg,23
2316 | _3Lb-lGfkL4_04621.jpg,23
2317 | ZODJivxqFtk_06133.jpg,23
2318 | BbOZCMbryI0_08238.jpg,23
2319 | C3qxIoyIysM_02008.jpg,171
2320 | JECooRTSJrI_03691.jpg,171
2321 | tuFjRSDwOjw_06284.jpg,97
2322 | cdPqGpij7vI_01975.jpg,97
2323 | bosCMWHBCmk_12288.jpg,171
2324 | G_XkHM8T2DQ_00758.jpg,98
2325 | W1rEZOyAfac_06936.jpg,171
2326 | D1YdfC6VFl8_06281.jpg,97
2327 | 20crtxQDlJw_02844.jpg,98
2328 | 4BWf4vQ3B9A_00323.jpg,171
2329 | fSZLEUFh5Zk_05966.jpg,98
2330 | IK-DRCL7YEc_03455.jpg,171
2331 | W1rEZOyAfac_00817.jpg,158
2332 | Srw7-xTmA2w_13763.jpg,171
2333 | fIBBWLjAFzU_04258.jpg,171
2334 | JThGq9_hSys_04513.jpg,0
2335 | 71bvKydh5gk_04295.jpg,66
2336 | 71bvKydh5gk_06046.jpg,66
2337 | 71bvKydh5gk_01894.jpg,66
2338 | 71bvKydh5gk_05964.jpg,66
2339 | 71bvKydh5gk_05121.jpg,66
2340 | 71bvKydh5gk_01177.jpg,66
2341 | 71bvKydh5gk_04650.jpg,66
2342 | a_HNXvnrbRU_01410.jpg,37
2343 | 71bvKydh5gk_01514.jpg,66
2344 | 71bvKydh5gk_03881.jpg,66
2345 | a_HNXvnrbRU_06140.jpg,0
2346 | JThGq9_hSys_06462.jpg,0
2347 | a_HNXvnrbRU_03681.jpg,66
2348 | R-gqVUPrkJQ_01528.jpg,72
2349 | R-gqVUPrkJQ_02621.jpg,72
2350 | wyuRF3R14CU_03343.jpg,175
2351 | 03FRktt6G1w_03775.jpg,175
2352 | b-ZmrXItjdo_00462.jpg,175
2353 | 21yKMd9GuqY_02629.jpg,175
2354 | wyuRF3R14CU_04492.jpg,175
2355 | wyuRF3R14CU_01264.jpg,175
2356 | raWjkVa0glM_00000.jpg,203
2357 | 21yKMd9GuqY_02404.jpg,175
2358 | 21yKMd9GuqY_02861.jpg,175
2359 | 03FRktt6G1w_00507.jpg,175
2360 | wyuRF3R14CU_01698.jpg,175
2361 | foUWwKuT080_05231.jpg,57
2362 | mIOI_3SQD_E_03727.jpg,57
2363 | g_PRZY0swPs_13873.jpg,57
2364 | 7cHw2fy9QHI_03376.jpg,4
2365 | SiHZxxBg0oU_00949.jpg,79
2366 | 2tlsmV7JfF8_04451.jpg,57
2367 | 7cHw2fy9QHI_03033.jpg,57
2368 | SiHZxxBg0oU_03856.jpg,2
2369 | mIOI_3SQD_E_04528.jpg,57
2370 | dYID8IVom6Y_01055.jpg,57
2371 | ZK84WqSF8SI_11458.jpg,128
2372 | oC6tnNPN0ng_07082.jpg,57
2373 | mDb_aFjbOEY_02481.jpg,57
2374 | i1-n-6frKgE_00942.jpg,57
2375 | 2UUTUSZjPLE_01161.jpg,57
2376 | fcNEifqy7uw_06585.jpg,57
2377 | jGy5eKKYiG0_05545.jpg,95
2378 | 2UUTUSZjPLE_01915.jpg,76
2379 | PDmTXM7vJ_g_08613.jpg,4
2380 | VunCqiW12_U_03781.jpg,57
2381 | FimPUrjMvP4_00428.jpg,57
2382 | DqPk86h7hCc_02343.jpg,128
2383 | Ytp-pm301Uo_06346.jpg,23
2384 | ooMslfBsJ0g_06446.jpg,57
2385 | k1FNu_Jy7CE_00720.jpg,128
2386 | ooMslfBsJ0g_05748.jpg,57
2387 | KIO0n_IjBaA_03378.jpg,57
2388 | pOggFBjZz_E_03125.jpg,57
2389 | foUWwKuT080_00000.jpg,57
2390 | Jjj8E3nUGSg_02328.jpg,4
2391 | foUWwKuT080_03930.jpg,190
2392 | bHg4v_zs144_11898.jpg,2
2393 | LCJtvtUlhPk_07454.jpg,100
2394 | hI2CHDTamx4_00317.jpg,204
2395 | LCJtvtUlhPk_07603.jpg,100
2396 | bHg4v_zs144_12218.jpg,2
2397 | 86fx2LzZw_I_06218.jpg,100
2398 | 5hCahYi63xQ_00828.jpg,38
2399 | 6nx__lXVHx0_00124.jpg,38
2400 | 7YgUMTGP9O8_02587.jpg,38
2401 | Bdx6xMuWQWU_01850.jpg,38
2402 | 9if1Dic7yX8_01195.jpg,38
2403 | v0K3u_NEZAY_01463.jpg,128
2404 | H41i4uOveUI_03350.jpg,38
2405 | 6Xp5Uf6Is1M_03860.jpg,38
2406 | 1G43kvVHd7o_07072.jpg,38
2407 | 6nx__lXVHx0_01333.jpg,38
2408 | 8S3xBVVdnLE_01745.jpg,38
2409 | 9IHbikWV-Wc_00527.jpg,38
2410 | AxDrPCnHq10_00324.jpg,38
2411 | RW8JgA-Qz8s_00831.jpg,38
2412 | 8Qo9yFFdTGo_04563.jpg,38
2413 | 1G43kvVHd7o_07683.jpg,38
2414 | GFI3-UaoyfQ_00581.jpg,38
2415 | aaMbyi0bgRk_02181.jpg,38
2416 | BvqcFo8yqU4_01880.jpg,38
2417 | 1G43kvVHd7o_02221.jpg,38
2418 | AxDrPCnHq10_06127.jpg,38
2419 | 1G43kvVHd7o_05227.jpg,38
2420 | K9QjlWQazr0_00576.jpg,38
2421 | ezzgZb-f58I_00770.jpg,38
2422 | ASuf8XVA1lk_02579.jpg,38
2423 | LxNI9xfwHJ8_00951.jpg,38
2424 | Alt-ehDc3fc_02257.jpg,38
2425 | 9OnP4yQUUjc_00981.jpg,38
2426 | 8UksnUyBfUA_00216.jpg,38
2427 | 6cBLBq8GqDM_03635.jpg,38
2428 | JFZuPJn9hCI_00517.jpg,38
2429 | 8S3xBVVdnLE_02451.jpg,38
2430 | ASuf8XVA1lk_00538.jpg,38
2431 | ASuf8XVA1lk_01546.jpg,38
2432 | 1Jq4tPutdGQ_02479.jpg,38
2433 | 1_JlhywUlvU_01797.jpg,38
2434 | 0_X2ScFTi5M_01888.jpg,38
2435 | 8Qo9yFFdTGo_00929.jpg,38
2436 | M6QJSbdmhsU_01705.jpg,38
2437 | 1G43kvVHd7o_01791.jpg,38
2438 | RW8JgA-Qz8s_02872.jpg,38
2439 | 1_JlhywUlvU_01128.jpg,38
2440 | -jNEYX7oKLI_09680.jpg,38
2441 | 7YgUMTGP9O8_02149.jpg,38
2442 | H41i4uOveUI_01182.jpg,38
2443 | 8UksnUyBfUA_02163.jpg,38
2444 | 1Jq4tPutdGQ_02652.jpg,38
2445 | H41i4uOveUI_01881.jpg,38
2446 | JJ6UIU14tRM_01874.jpg,38
2447 | 3RJayxWw-jE_01415.jpg,38
2448 | JFZuPJn9hCI_00404.jpg,38
2449 | AdX-7rhYR6M_01083.jpg,38
2450 | 1Jq4tPutdGQ_01010.jpg,38
2451 | -h9uTe3B0SM_02937.jpg,38
2452 | M6QJSbdmhsU_03760.jpg,38
2453 | LxNI9xfwHJ8_00310.jpg,38
2454 | NVeTGvwk8y0_00682.jpg,38
2455 | LxNI9xfwHJ8_01985.jpg,38
2456 | 6Xp5Uf6Is1M_00377.jpg,38
2457 | LufrnOktZiA_02291.jpg,38
2458 | REmAcVnNYbE_05363.jpg,38
2459 | 6cBLBq8GqDM_04070.jpg,38
2460 | 0JrNDWfDZu0_00000.jpg,38
2461 | REmAcVnNYbE_08822.jpg,38
2462 | Alt-ehDc3fc_03012.jpg,38
2463 | S6R8EzKqRQI_02812.jpg,38
2464 | ASuf8XVA1lk_02833.jpg,38
2465 | KftPwi-VtOI_00759.jpg,38
2466 | 9if1Dic7yX8_00290.jpg,38
2467 | 8QFBUkOrLN4_01217.jpg,38
2468 | 6Xp5Uf6Is1M_02179.jpg,38
2469 | BvqcFo8yqU4_02147.jpg,38
2470 | ezzgZb-f58I_01003.jpg,38
2471 | Bdx6xMuWQWU_01259.jpg,38
2472 | 1G43kvVHd7o_07659.jpg,38
2473 | vuUULRIHA24_00685.jpg,95
2474 | M6QJSbdmhsU_01931.jpg,38
2475 | 1G43kvVHd7o_07331.jpg,38
2476 | 1G43kvVHd7o_01536.jpg,38
2477 | 9q4WjjSxD68_01564.jpg,38
2478 | M6QJSbdmhsU_00884.jpg,38
2479 | 6Xp5Uf6Is1M_02050.jpg,0
2480 | 9OnP4yQUUjc_00198.jpg,38
2481 | JFZuPJn9hCI_03074.jpg,38
2482 | AxDrPCnHq10_03110.jpg,38
2483 | 9q4WjjSxD68_01908.jpg,38
2484 | mRTG7LPH67M_09301.jpg,38
2485 | 7yJpVFzSteU_01847.jpg,38
2486 | 59sSNNykfpQ_00101.jpg,38
2487 | M6QJSbdmhsU_02345.jpg,38
2488 | 59sSNNykfpQ_01333.jpg,38
2489 | 6nx__lXVHx0_00248.jpg,38
2490 | Bdx6xMuWQWU_01094.jpg,38
2491 | Al6KQ3lXAXs_00348.jpg,38
2492 | K9QjlWQazr0_01310.jpg,38
2493 | 6cBLBq8GqDM_01415.jpg,38
2494 | FY--KGLMd5M_00238.jpg,38
2495 | 6nx__lXVHx0_00544.jpg,38
2496 | -h9uTe3B0SM_02573.jpg,38
2497 | REmAcVnNYbE_08279.jpg,38
2498 | 5A3fNXvkZqg_05549.jpg,38
2499 | JFZuPJn9hCI_03667.jpg,38
2500 | 6Xp5Uf6Is1M_03605.jpg,38
2501 | 6nx__lXVHx0_00832.jpg,38
2502 | KftPwi-VtOI_04345.jpg,38
2503 | H41i4uOveUI_02600.jpg,38
2504 | AdX-7rhYR6M_01445.jpg,38
2505 | C9zi1ILvizc_00407.jpg,135
2506 | RW8JgA-Qz8s_00349.jpg,38
2507 | K9QjlWQazr0_01164.jpg,38
2508 | H41i4uOveUI_04554.jpg,38
2509 | 6cBLBq8GqDM_00809.jpg,38
2510 | LxNI9xfwHJ8_01807.jpg,38
2511 | aaMbyi0bgRk_02857.jpg,38
2512 | sqIyA9sGZHI_02380.jpg,17
2513 | UN0SLzFNpDs_09359.jpg,17
2514 | nuxM5-PsQ-Q_01072.jpg,190
2515 | nuxM5-PsQ-Q_01898.jpg,17
2516 | LWAg6ACqbLU_02951.jpg,41
2517 | VV1cMmCKxmY_04593.jpg,0
2518 | GTCGzrqE_BA_02755.jpg,17
2519 | U3YQMfV3dkQ_05317.jpg,17
2520 | yaoES9T8pTk_06114.jpg,128
2521 | XigySGqneLc_06764.jpg,17
2522 | oV7h8mu4-lw_01552.jpg,100
2523 | ya_Qj50LylI_00510.jpg,137
2524 | jcmeJRw3C4g_03678.jpg,137
2525 | oDQUqtvQMOM_01078.jpg,0
2526 | DkYHNO2vNG4_01525.jpg,17
2527 | nuxM5-PsQ-Q_03280.jpg,135
2528 | 5IpFaVhRPAE_07609.jpg,17
2529 | 7AjmaDsauVY_03032.jpg,17
2530 | XigySGqneLc_04466.jpg,17
2531 | kchEKv7f7jo_01105.jpg,17
2532 | 5X6e7Su9CdQ_08026.jpg,18
2533 | cd_wozV-akk_05204.jpg,44
2534 | hMbDxERftcs_06684.jpg,44
2535 | BZ6isBVPcpc_05088.jpg,44
2536 | cd_wozV-akk_02834.jpg,44
2537 | BZ6isBVPcpc_00768.jpg,44
2538 | QaLmTRpFe3Y_13479.jpg,44
2539 | QaLmTRpFe3Y_13443.jpg,44
2540 | 4WmcYel7e70_06220.jpg,44
2541 | QaLmTRpFe3Y_13463.jpg,44
2542 | tz_rocCyfho_01928.jpg,44
2543 | 1xF7E7-cV48_05645.jpg,44
2544 | hMbDxERftcs_07060.jpg,44
2545 | BZ6isBVPcpc_04720.jpg,44
2546 | PDp_ev8oirQ_01159.jpg,44
2547 | hMbDxERftcs_01990.jpg,44
2548 | 15qzhJiCXkA_04540.jpg,44
2549 | P3VxJ-usb7Q_00499.jpg,30
2550 | oxwSmoIZ2aU_01271.jpg,44
2551 | QaLmTRpFe3Y_13415.jpg,44
2552 | 3o3rX3eMcj0_06735.jpg,44
2553 | UMbA6-vew8s_00958.jpg,44
2554 | UMbA6-vew8s_03344.jpg,44
2555 | hMbDxERftcs_06784.jpg,44
2556 | 56j_3cYjpwY_03116.jpg,44
2557 | hMbDxERftcs_01757.jpg,44
2558 | hMbDxERftcs_03268.jpg,44
2559 | UMbA6-vew8s_00781.jpg,44
2560 | BZ6isBVPcpc_02713.jpg,44
2561 | hMbDxERftcs_06751.jpg,44
2562 | 1SzDPmAOZKo_12001.jpg,11
2563 | BZ6isBVPcpc_06344.jpg,44
2564 | BZ6isBVPcpc_03364.jpg,44
2565 | BZ6isBVPcpc_06588.jpg,44
2566 | cd_wozV-akk_05313.jpg,44
2567 | BZ6isBVPcpc_06528.jpg,44
2568 | 56j_3cYjpwY_02895.jpg,44
2569 | 6uWjAoRVE38_12197.jpg,44
2570 | cd_wozV-akk_03047.jpg,44
2571 | 56j_3cYjpwY_02590.jpg,44
2572 | hMbDxERftcs_03670.jpg,44
2573 | BZ6isBVPcpc_06435.jpg,44
2574 | BZ6isBVPcpc_01802.jpg,44
2575 | KXk1yI9sdDE_02109.jpg,44
2576 | QaLmTRpFe3Y_12598.jpg,44
2577 | QaLmTRpFe3Y_13578.jpg,44
2578 | O-y4EsSTXKE_01759.jpg,44
2579 | QaLmTRpFe3Y_13708.jpg,44
2580 | O-y4EsSTXKE_01281.jpg,44
2581 | hMbDxERftcs_06737.jpg,44
2582 | BZ6isBVPcpc_05219.jpg,44
2583 | QaLmTRpFe3Y_13613.jpg,44
2584 | oxwSmoIZ2aU_01175.jpg,44
2585 | 1SzDPmAOZKo_02542.jpg,128
2586 | O-y4EsSTXKE_01558.jpg,44
2587 | 15qzhJiCXkA_04818.jpg,140
2588 | 56j_3cYjpwY_02472.jpg,44
2589 | oxwSmoIZ2aU_01215.jpg,44
2590 | oxwSmoIZ2aU_01260.jpg,44
2591 | O-y4EsSTXKE_01069.jpg,44
2592 | hMbDxERftcs_06915.jpg,44
2593 | BZ6isBVPcpc_04825.jpg,44
2594 | UMbA6-vew8s_01013.jpg,44
2595 | 5KPBSwNOVPM_02986.jpg,44
2596 | BZ6isBVPcpc_05976.jpg,44
2597 | ltz1FX40R6Y_00604.jpg,44
2598 | QaLmTRpFe3Y_01449.jpg,44
2599 | QaLmTRpFe3Y_13609.jpg,44
2600 | tz_rocCyfho_01580.jpg,44
2601 | cd_wozV-akk_03113.jpg,44
2602 | BZ6isBVPcpc_05128.jpg,44
2603 | QaLmTRpFe3Y_13742.jpg,44
2604 | KXk1yI9sdDE_02679.jpg,44
2605 | BZ6isBVPcpc_00826.jpg,44
2606 | O-y4EsSTXKE_05061.jpg,44
2607 | BZ6isBVPcpc_04024.jpg,44
2608 | hMbDxERftcs_07069.jpg,44
2609 | oxwSmoIZ2aU_00923.jpg,44
2610 | BZ6isBVPcpc_02652.jpg,44
2611 | 1F1RwxWAR4I_00030.jpg,7
2612 | QaLmTRpFe3Y_13651.jpg,44
2613 | BZ6isBVPcpc_06310.jpg,44
2614 | 56j_3cYjpwY_06108.jpg,44
2615 | O-y4EsSTXKE_04702.jpg,44
2616 | 56j_3cYjpwY_04053.jpg,44
2617 | O-y4EsSTXKE_01480.jpg,44
2618 | BZ6isBVPcpc_04837.jpg,44
2619 | BZ6isBVPcpc_06728.jpg,44
2620 | KAc78Q9B46g_01182.jpg,44
2621 | BZ6isBVPcpc_02757.jpg,44
2622 | O-y4EsSTXKE_01225.jpg,44
2623 | UMbA6-vew8s_03359.jpg,44
2624 | QaLmTRpFe3Y_13693.jpg,44
2625 | UMbA6-vew8s_00644.jpg,44
2626 | O-y4EsSTXKE_01549.jpg,44
2627 | P3VxJ-usb7Q_00060.jpg,44
2628 | O-y4EsSTXKE_04618.jpg,44
2629 | O-y4EsSTXKE_04687.jpg,44
2630 | hMbDxERftcs_07245.jpg,44
2631 | 56j_3cYjpwY_05341.jpg,44
2632 | cd_wozV-akk_06076.jpg,44
2633 | cd_wozV-akk_05802.jpg,44
2634 | BZ6isBVPcpc_03760.jpg,44
2635 | QaLmTRpFe3Y_13704.jpg,44
2636 | sOiZGl_1l8c_04259.jpg,137
2637 | _RY8A8016AI_03008.jpg,137
2638 | VUiQQ04F3Bg_03937.jpg,137
2639 | Gp6cl6nhKNo_01721.jpg,137
2640 | UaKA8bkm718_00211.jpg,137
2641 | GnaLVgY4OYs_00866.jpg,156
2642 | sOiZGl_1l8c_01570.jpg,137
2643 | EzFZ9MG6Wow_03548.jpg,137
2644 | kiicywUWz34_01920.jpg,137
2645 | _RY8A8016AI_05125.jpg,137
2646 | Sdsic3d8XN8_03628.jpg,137
2647 | Sdsic3d8XN8_04428.jpg,137
2648 | qZVhbzBCd2w_04075.jpg,137
2649 | Rn71Eyl3-mU_07723.jpg,137
2650 | qZVhbzBCd2w_01652.jpg,137
2651 | JmktV9YpeI8_04065.jpg,137
2652 | VUiQQ04F3Bg_01875.jpg,137
2653 | QXfgYjqd4TI_17000.jpg,137
2654 | 5-gJUpK7Gyk_00185.jpg,137
2655 | Rn71Eyl3-mU_02839.jpg,137
2656 | VUiQQ04F3Bg_04060.jpg,137
2657 | JmktV9YpeI8_03213.jpg,137
2658 | EzFZ9MG6Wow_01221.jpg,137
2659 | Ono_hs229-w_13000.jpg,137
2660 | _RY8A8016AI_01026.jpg,137
2661 | qZVhbzBCd2w_02097.jpg,137
2662 | VUiQQ04F3Bg_02214.jpg,137
2663 | VLdG2ZwZiQw_00000.jpg,128
2664 | ku4PgCRMgRo_01670.jpg,20
2665 | 8oEJBAtayZA_00000.jpg,20
2666 | HD-fX4-rEgc_15107.jpg,7
2667 | jW_GekniXSQ_01816.jpg,20
2668 | 2fajzbgwvxg_02704.jpg,20
2669 | ku4PgCRMgRo_01773.jpg,20
2670 | jW_GekniXSQ_00654.jpg,20
2671 | 2SjB3RXZlJw_09325.jpg,20
2672 | GNLHOuMyTqk_01126.jpg,20
2673 | ku4PgCRMgRo_01581.jpg,20
2674 | FLbxYOfTqZA_02441.jpg,20
2675 | 525fL0YlGL8_00770.jpg,20
2676 | 525fL0YlGL8_00749.jpg,20
2677 | i0dyBugWcjk_00723.jpg,20
2678 | FLbxYOfTqZA_02155.jpg,20
2679 | hrd6ce3ZBig_01402.jpg,15
2680 | 0aAnMnAYzK4_00599.jpg,128
2681 | 0xrMSFLOcT8_00784.jpg,20
2682 | IElvQVYPmYA_01604.jpg,20
2683 | FLbxYOfTqZA_02147.jpg,20
2684 | vV3eVPiO_ao_02524.jpg,20
2685 | 0gMnsxzo2Do_00777.jpg,89
2686 | gvh7GOfUh2M_01525.jpg,8
2687 | FLbxYOfTqZA_01884.jpg,20
2688 | WRYsx28GqBk_01878.jpg,128
2689 | 0gMnsxzo2Do_02864.jpg,113
2690 | CCQ3VzbdO8k_01025.jpg,97
2691 | ku4PgCRMgRo_00571.jpg,20
2692 | 1Kk80h_hob8_03947.jpg,20
2693 | 2SjB3RXZlJw_04860.jpg,20
2694 | GNLHOuMyTqk_02352.jpg,20
2695 | w0yl4A_FwsE_00874.jpg,20
2696 | SsIzsZstHwE_02416.jpg,15
2697 | dfF8Xu_kESE_03329.jpg,20
2698 | Q9g3seOtyN0_00165.jpg,135
2699 | ku4PgCRMgRo_02828.jpg,20
2700 | GNLHOuMyTqk_02284.jpg,20
2701 | B8uBvSbiIfs_02314.jpg,20
2702 | i0dyBugWcjk_02627.jpg,20
2703 | IElvQVYPmYA_00252.jpg,10
2704 | 0xrMSFLOcT8_00850.jpg,20
2705 | FLbxYOfTqZA_02391.jpg,20
2706 | SsIzsZstHwE_01537.jpg,135
2707 | gGxHm_QarrM_02284.jpg,20
2708 | 525fL0YlGL8_00798.jpg,20
2709 | hrd6ce3ZBig_03871.jpg,20
2710 | jW_GekniXSQ_01118.jpg,20
2711 | JLM905gSDH0_03550.jpg,20
2712 | FLbxYOfTqZA_01929.jpg,20
2713 | i0dyBugWcjk_02736.jpg,20
2714 | FLbxYOfTqZA_01648.jpg,20
2715 | Km65akBPivw_00029.jpg,20
2716 | 5EWgJtWdKuE_07814.jpg,20
2717 | xKQVguroRHI_01788.jpg,128
2718 | ku4PgCRMgRo_02009.jpg,20
2719 | i0dyBugWcjk_01166.jpg,20
2720 | Mb9i5PLBffY_00694.jpg,128
2721 | MoYNXtF5dcs_05830.jpg,20
2722 | vV3eVPiO_ao_01010.jpg,20
2723 | i0dyBugWcjk_01256.jpg,20
2724 | gdCMixslLss_00335.jpg,128
2725 | vV3eVPiO_ao_02225.jpg,20
2726 | 29zj0hAcSRY_02520.jpg,20
2727 | 2fajzbgwvxg_01281.jpg,111
2728 | hWukFEFUyiM_00313.jpg,44
2729 | i0dyBugWcjk_00420.jpg,20
2730 | WRYsx28GqBk_06422.jpg,128
2731 | 5EWgJtWdKuE_07998.jpg,20
2732 | ku4PgCRMgRo_02963.jpg,20
2733 | WRYsx28GqBk_01721.jpg,128
2734 | 525fL0YlGL8_03423.jpg,20
2735 | 88Ai4_KIsTY_06813.jpg,51
2736 | yUfRZzJ4zsA_00290.jpg,13
2737 | e-EJb4XsT68_06756.jpg,13
2738 | GglZ1qQQB18_08930.jpg,13
2739 | 88Ai4_KIsTY_03063.jpg,0
2740 | YGtroBi88P4_03042.jpg,79
2741 | rzHP6-i2u9Q_03251.jpg,26
2742 | QWfD6XdbJdc_04582.jpg,158
2743 | LB6RC356-XY_04504.jpg,13
2744 | 5yja_52nQZY_11748.jpg,8
2745 | QWfD6XdbJdc_04473.jpg,158
2746 | KFP5ZnveEbg_01183.jpg,2
2747 | LB6RC356-XY_05371.jpg,94
2748 | zduSWebImZs_05611.jpg,13
2749 | JxtqlYeF4hU_03370.jpg,13
2750 | OHtlTKZUf4o_04138.jpg,13
2751 | u3XwMdyia6c_02341.jpg,13
2752 | PfI2NA10lDQ_07526.jpg,24
2753 | 0uBOtQOO70Y_05301.jpg,53
2754 | ynUjo99Gzbk_06076.jpg,53
2755 | ynUjo99Gzbk_12190.jpg,53
2756 | 0uBOtQOO70Y_10104.jpg,53
2757 | L2zqWf_m9DA_11623.jpg,53
2758 | 0uBOtQOO70Y_10340.jpg,53
2759 | ynUjo99Gzbk_02381.jpg,53
2760 | t-UNRLzXAQc_03538.jpg,53
2761 | t-UNRLzXAQc_05266.jpg,53
2762 | t-UNRLzXAQc_02434.jpg,53
2763 | ynUjo99Gzbk_06210.jpg,53
2764 | ynUjo99Gzbk_02823.jpg,53
2765 | ynUjo99Gzbk_12335.jpg,53
2766 | 0uBOtQOO70Y_10365.jpg,53
2767 | 3YPl_BcQtmw_02181.jpg,53
2768 | zvUybiBPJHs_00203.jpg,53
2769 | L2zqWf_m9DA_04153.jpg,53
2770 | t-UNRLzXAQc_02585.jpg,53
2771 | TUKvVJs3iTw_02183.jpg,53
2772 | 2B9PUiVL0cc_02460.jpg,53
2773 | rQRbDVG-ra0_01752.jpg,9
2774 | Shg_lkccrr4_00000.jpg,9
2775 | RmgbuPKg__U_03561.jpg,9
2776 | asrVJKZJWnI_03026.jpg,0
2777 | EcEG4d4E1-E_00401.jpg,9
2778 | coIewQpqTuU_00148.jpg,9
2779 | RCun6SXJoBY_00870.jpg,9
2780 | ecU_IzHfpXo_03211.jpg,9
2781 | TOo51nhMoVw_00000.jpg,9
2782 | zwN4tqQrFkA_00304.jpg,9
2783 | 5fIeFJxCM1A_03144.jpg,9
2784 | Flt3iCWbgtw_00767.jpg,9
2785 | 2yTxvDQpcL4_01989.jpg,9
2786 | zwN4tqQrFkA_00000.jpg,9
2787 | 3BiYQbD-mVk_04328.jpg,9
2788 | JH1P4ma_BN8_01756.jpg,13
2789 | rQRbDVG-ra0_02117.jpg,9
2790 | lg5__mXMI6k_09684.jpg,9
2791 | LGr0AamJ7B0_00870.jpg,128
2792 | dOn5DD8CJNg_00237.jpg,9
2793 | zmTvvA2Klrk_03324.jpg,9
2794 | 9VY9Xp2ZhK8_03613.jpg,9
2795 | IZ63T51-CXk_03517.jpg,9
2796 | 2x4p5dW8vfE_00712.jpg,9
2797 | JH1P4ma_BN8_01653.jpg,9
2798 | ERcrtUpyxIE_00000.jpg,9
2799 | gJJlfUv9YfU_00000.jpg,9
2800 | 2yTxvDQpcL4_00489.jpg,9
2801 | 60MTp_r99P8_00651.jpg,9
2802 | lg5__mXMI6k_06746.jpg,9
2803 | MShircY-SPo_01169.jpg,128
2804 | 6eSlsyTWu08_03565.jpg,9
2805 | 2x4p5dW8vfE_04904.jpg,9
2806 | 5fIeFJxCM1A_03038.jpg,19
2807 | XjPHNaQuOCM_00370.jpg,9
2808 | 7fIT4dksjHo_00404.jpg,9
2809 | ZZrqTcRenok_00846.jpg,9
2810 | 60MTp_r99P8_00000.jpg,82
2811 | RmgbuPKg__U_12604.jpg,9
2812 | a1VwE0_QERQ_01889.jpg,9
2813 | AT7p0xgZlG0_04003.jpg,9
2814 | a1VwE0_QERQ_03329.jpg,9
2815 | AcbUmQKf8u0_14679.jpg,7
2816 | xVS66VJQuZs_08491.jpg,7
2817 | CDFbfUnr2lE_08814.jpg,171
2818 | 3WKPWJpPCwo_01066.jpg,7
2819 | qMpLUNbdlVY_01813.jpg,7
2820 | G6UPds6ZABo_00558.jpg,7
2821 | jlF_ekXp5PA_04379.jpg,44
2822 | DSG1dxCLyyY_13191.jpg,7
2823 | G6UPds6ZABo_12088.jpg,7
2824 | AcbUmQKf8u0_10232.jpg,7
2825 | K62-WvsPUjg_02293.jpg,8
2826 | Zfw2-lEUoOo_08444.jpg,18
2827 | rcg9bCiAeMw_08937.jpg,7
2828 | xalnp-M9yos_04689.jpg,135
2829 | 3WKPWJpPCwo_00932.jpg,7
2830 | muV7h3iwbK0_05457.jpg,7
2831 | HS3pfBwBlcM_00317.jpg,7
2832 | V_m01zLjKEM_01033.jpg,11
2833 | pslzMze9rNk_07835.jpg,7
2834 | G6UPds6ZABo_15660.jpg,7
2835 | AcbUmQKf8u0_10395.jpg,7
2836 | e1kn_f2RHYc_09339.jpg,7
2837 | QjZ1J7kNs8I_05524.jpg,7
2838 | RO_Myj1_6_M_00295.jpg,94
2839 | 8nhFfPdV8KA_00482.jpg,7
2840 | EKVprq89x9o_00493.jpg,7
2841 | 7pKyheM4YrE_00067.jpg,128
2842 | TS8IFyVHcFU_05093.jpg,7
2843 | NgEvA7lgAXg_01850.jpg,98
2844 | 3WKPWJpPCwo_00820.jpg,7
2845 | o3DEA9Uc-lg_05214.jpg,7
2846 | AcbUmQKf8u0_14512.jpg,7
2847 | s_5lbrEMZDs_00768.jpg,7
2848 | xoB_acwcu5s_02137.jpg,7
2849 | eBc2rGFeT74_01617.jpg,7
2850 | _SA1Zftf26k_02229.jpg,7
2851 | uLf-qmuWKfo_12654.jpg,7
2852 | ZSkQdBGD47Q_06201.jpg,11
2853 | 3WKPWJpPCwo_00781.jpg,7
2854 | zPCjHyw27Yg_05559.jpg,9
2855 | 6QSuzQl6pn4_02965.jpg,7
2856 | tqBkcXJ_tFw_03985.jpg,7
2857 | WuBNsSc1O0E_02430.jpg,7
2858 | eBc2rGFeT74_01087.jpg,7
2859 | G6UPds6ZABo_00979.jpg,7
2860 | MupYgZ_edEE_03716.jpg,11
2861 | TKFguFlDy5U_00915.jpg,7
2862 | ke3VIGEmdqo_03152.jpg,7
2863 | b6YGmXBYj1c_01453.jpg,7
2864 | cGypZ9mJLfM_11156.jpg,7
2865 | ES142HwbyDc_00261.jpg,7
2866 | SuIgG4UY8zk_00468.jpg,7
2867 | NbFr4WC1erk_07900.jpg,7
2868 | Wjrsqh7dcYc_04906.jpg,7
2869 | bKUAARS2TRg_02722.jpg,22
2870 | 7TTKuLq2LUE_11038.jpg,22
2871 | VWYaGgF7TYg_01522.jpg,22
2872 | VWYaGgF7TYg_06935.jpg,22
2873 | 0Jqq9NH7QQE_00317.jpg,19
2874 | VWYaGgF7TYg_02656.jpg,22
2875 | 2CJs7kOG6Gk_03924.jpg,22
2876 | FzTu3LdaitQ_00377.jpg,22
2877 | VWYaGgF7TYg_06927.jpg,22
2878 | VWYaGgF7TYg_04734.jpg,22
2879 | Y0WqKx7fDb8_06225.jpg,10
2880 | VWYaGgF7TYg_06620.jpg,22
2881 | VWYaGgF7TYg_02639.jpg,22
2882 | Gj2K7RfKObY_04866.jpg,128
2883 | c5-13T8sZ-8_03740.jpg,28
2884 | TlntUtxh6Yw_00239.jpg,7
2885 | E8mt-qt4r-U_00284.jpg,114
2886 | NbM3kgRaf0w_00036.jpg,114
2887 | P0ppQdhfWgc_04834.jpg,114
2888 | TMLGBzwXsqg_02974.jpg,114
2889 | NbM3kgRaf0w_00070.jpg,114
2890 | IzY8y0o8Wc8_03944.jpg,114
2891 | cuJRm6N0SAo_02132.jpg,114
2892 | lxRrReZJqHY_02139.jpg,1
2893 | mgeXiV3rF0w_00176.jpg,114
2894 | o-Kgl3ePAoo_00436.jpg,5
2895 | I_a3HLdwiZ8_00000.jpg,168
2896 | s3HhNlysFDs_01570.jpg,5
2897 | CxpQmISI0TA_02239.jpg,5
2898 | J2B0iR2VSPg_06992.jpg,5
2899 | s3HhNlysFDs_01699.jpg,5
2900 | GLAVw2dfoeA_03308.jpg,5
2901 | MW2rvYkY1qM_01035.jpg,5
2902 | lLy67c7_PPY_08863.jpg,5
2903 | YXtUetabb30_00317.jpg,5
2904 | 82jhP7g49UI_00112.jpg,5
2905 | J2B0iR2VSPg_02086.jpg,5
2906 | J2B0iR2VSPg_05214.jpg,5
2907 | MW2rvYkY1qM_01627.jpg,5
2908 | En_tBwfm4d8_01666.jpg,135
2909 | J2B0iR2VSPg_05014.jpg,5
2910 | 6gMcFUt4SiE_03053.jpg,0
2911 | vow-kxTPatc_15645.jpg,2
2912 | E0lvSZydTnU_06213.jpg,95
2913 | Sw1mqdKibX8_00319.jpg,0
2914 | URM04TnbIr4_01009.jpg,95
2915 | E0lvSZydTnU_00267.jpg,95
2916 | fRf1Lge4KBQ_01556.jpg,13
2917 | 6gMcFUt4SiE_01778.jpg,95
2918 | E0lvSZydTnU_05824.jpg,95
2919 | ChEtlFxGq5M_00774.jpg,95
2920 | URM04TnbIr4_01820.jpg,95
2921 | M7t0b_9nTcA_00809.jpg,95
2922 | BWc7LycDCkM_03078.jpg,95
2923 | E0lvSZydTnU_03096.jpg,95
2924 | E0lvSZydTnU_03652.jpg,95
2925 | M7t0b_9nTcA_04625.jpg,95
2926 | E0lvSZydTnU_04917.jpg,95
2927 | URM04TnbIr4_02270.jpg,2
2928 | 6K3KKTCpxBE_03679.jpg,140
2929 | 15qzhJiCXkA_02001.jpg,140
2930 | 6K3KKTCpxBE_02554.jpg,140
2931 | iMGUeElyj6Q_00583.jpg,10
2932 | jeh2GWO93n0_00394.jpg,128
2933 | xECHntI5dmA_00158.jpg,128
2934 | bN4dwJ6vjIo_08837.jpg,64
2935 | bN4dwJ6vjIo_07482.jpg,26
2936 | g3Vv6qFne5Y_00457.jpg,135
2937 | g3Vv6qFne5Y_02806.jpg,135
2938 | vuUULRIHA24_03323.jpg,4
2939 | WeuNWwzw94Y_07222.jpg,4
2940 | vuUULRIHA24_03106.jpg,4
2941 | WeuNWwzw94Y_07441.jpg,4
2942 | vuUULRIHA24_00849.jpg,4
2943 | vuUULRIHA24_03434.jpg,4
2944 | WeuNWwzw94Y_07198.jpg,4
2945 | WeuNWwzw94Y_07242.jpg,4
2946 | vuUULRIHA24_02116.jpg,4
2947 | vuUULRIHA24_02061.jpg,4
2948 | vuUULRIHA24_02239.jpg,4
2949 | WeuNWwzw94Y_07469.jpg,4
2950 | vuUULRIHA24_03403.jpg,4
2951 | vuUULRIHA24_03487.jpg,4
2952 | qK4UqCYUGqg_00541.jpg,82
2953 | EiDMExRyqXM_00015.jpg,82
2954 | qcfgrlUsTF8_01812.jpg,82
2955 | zf_jkkDh6fQ_02001.jpg,94
2956 | t-dE3HaGzyE_00723.jpg,82
2957 | S3uNZPdV8c8_00052.jpg,82
2958 | AW62Hh3_tDc_01922.jpg,82
2959 | EiDMExRyqXM_00096.jpg,82
2960 | TqxZS009jek_00060.jpg,82
2961 | qK4UqCYUGqg_00138.jpg,82
2962 | vhKcgK0itEI_01345.jpg,82
2963 | gspQNjbv7QI_00000.jpg,82
2964 | xlnWN2PKjtw_00707.jpg,82
2965 | R-F0kmif_vM_00601.jpg,82
2966 | Xb7tEXZ0Y9E_04040.jpg,82
2967 | x2H7zcqjplc_01338.jpg,7
2968 | dwlBZCI6PoY_00783.jpg,82
2969 | E9c7ExiRn7g_01233.jpg,95
2970 | titK5hztzAo_01157.jpg,94
2971 | Vw-nYY39hkc_01487.jpg,82
2972 | qK4UqCYUGqg_00440.jpg,82
2973 | qcfgrlUsTF8_00119.jpg,82
2974 | n3tHAnONtno_00343.jpg,82
2975 | OhJ3JxwdRss_01863.jpg,82
2976 | rImRX01WUsc_02472.jpg,82
2977 | 3F2ZXUVYW7k_01721.jpg,82
2978 | 7vMvnOEzzHo_01273.jpg,82
2979 | xUYT9x8VmbU_00221.jpg,82
2980 | rImRX01WUsc_03643.jpg,82
2981 | HbmzR0TdFXc_06299.jpg,120
2982 | OP0AVKxDw0c_00309.jpg,190
2983 | m400_PEJZSM_04931.jpg,151
2984 | HnMeRvVY6qY_02742.jpg,151
2985 | YoOs-sMH0Uc_03921.jpg,190
2986 | dO-leAHJFmk_09946.jpg,53
2987 | YN8IXENqRgo_06504.jpg,190
2988 | IU4iexw75sA_01171.jpg,190
2989 | m400_PEJZSM_05629.jpg,190
2990 | YoOs-sMH0Uc_01126.jpg,190
2991 | V6ukiO5e5wY_02465.jpg,190
2992 | V6ukiO5e5wY_00306.jpg,190
2993 | IU4iexw75sA_01712.jpg,190
2994 | m400_PEJZSM_10510.jpg,190
2995 | WZj8tTFdUVI_03053.jpg,27
2996 | IU4iexw75sA_01293.jpg,190
2997 | V11qLE5ZkhM_00841.jpg,190
2998 | ubs38mhIPM0_01743.jpg,190
2999 | Y1EVFM-ajTk_00916.jpg,190
3000 | YN8IXENqRgo_07333.jpg,190
3001 | 6JYAiTys2XA_02125.jpg,190
3002 | m400_PEJZSM_13175.jpg,190
3003 | I06jLnhThEM_00252.jpg,41
3004 | I06jLnhThEM_01693.jpg,41
3005 | U3YQMfV3dkQ_06452.jpg,41
3006 | RJ8jdf6A4KQ_04589.jpg,41
3007 | I06jLnhThEM_00773.jpg,17
3008 | I06jLnhThEM_01324.jpg,41
3009 | t6QszhPKU-M_00046.jpg,150
3010 | HruBZKH7r9I_06575.jpg,30
3011 | MeErpy5tF8I_03402.jpg,150
3012 | GQLnPBsXQpM_12551.jpg,150
3013 | KqcWVJ7l47g_01408.jpg,150
3014 | 9m4i1zJDPdU_00000.jpg,56
3015 | DGrDF0ScE5I_10717.jpg,150
3016 | Ifv1WD18rjM_08476.jpg,150
3017 | HruBZKH7r9I_05967.jpg,135
3018 | NVIgqAbZJJc_10390.jpg,150
3019 | CgbsjsldpMs_05990.jpg,56
3020 | Ifv1WD18rjM_08480.jpg,150
3021 | AmKWUbgZW9Q_02838.jpg,150
3022 | 9ciiHYmscGY_00567.jpg,13
3023 | u2iO55GnsCs_13300.jpg,150
3024 | GQLnPBsXQpM_05294.jpg,150
3025 | KqcWVJ7l47g_06327.jpg,150
3026 | 7byhPYZNgKs_16593.jpg,150
3027 | KqcWVJ7l47g_08280.jpg,150
3028 | HruBZKH7r9I_07342.jpg,30
3029 | 5Dle76vhHNM_02222.jpg,97
3030 | hMnu4nY8Ip8_02165.jpg,97
3031 | AeaBnPlp32M_00831.jpg,50
3032 | AeaBnPlp32M_00728.jpg,50
3033 | P97ByQnE4D0_02130.jpg,50
3034 | bWOHLPBlGtE_00363.jpg,50
3035 | TP22bO95WuE_04727.jpg,97
3036 | DrpNgJU3lwI_07738.jpg,97
3037 | ecrCsNOmSJk_03132.jpg,50
3038 | XBIe9umQJv0_04904.jpg,50
3039 | Px8L_VXWU9Y_02814.jpg,50
3040 | 5Dle76vhHNM_03252.jpg,97
3041 | 8Qk8-TNi7NM_03144.jpg,50
3042 | FRA5wEek8Co_07230.jpg,50
3043 | 9OhlQrG_G7E_00179.jpg,50
3044 | FRA5wEek8Co_07458.jpg,50
3045 | FRA5wEek8Co_00476.jpg,24
3046 | 5Dle76vhHNM_03231.jpg,50
3047 | WQrZLu4uVbE_00928.jpg,8
3048 | 5BZh3OCuoS0_02981.jpg,97
3049 | FINA9imlhoU_02856.jpg,50
3050 | 4ckVVmPvuZI_09008.jpg,50
3051 | a3HK1Oyfgd8_07316.jpg,50
3052 | 5Dle76vhHNM_03948.jpg,50
3053 | PGLSOH1iXEM_00604.jpg,50
3054 | 9OhlQrG_G7E_01898.jpg,8
3055 | 9c2Atsc4JnM_04322.jpg,50
3056 | hMnu4nY8Ip8_03311.jpg,97
3057 | jW_GekniXSQ_09516.jpg,28
3058 | jW_GekniXSQ_09427.jpg,28
3059 | 6Joey7jIQio_04961.jpg,11
3060 | jW_GekniXSQ_09417.jpg,28
3061 | 3WKPWJpPCwo_00043.jpg,30
3062 | 3Jqx7d4gb4I_00751.jpg,7
3063 | KS7oTEqrQtA_01197.jpg,30
3064 | 2B_RbYc-GEE_01367.jpg,30
3065 | P-RHswHbnow_00761.jpg,30
3066 | KS7oTEqrQtA_00911.jpg,30
3067 | P-RHswHbnow_04178.jpg,135
3068 | ZWTofuyfZZs_03350.jpg,27
3069 | 2B_RbYc-GEE_01399.jpg,30
3070 | LZ9KXy8G0Lg_00000.jpg,120
3071 | 6pJ6P5HGLIg_00441.jpg,30
3072 | kTep0iQTuS8_06274.jpg,30
3073 | cRBtP18SL3U_07559.jpg,30
3074 | 6pJ6P5HGLIg_00358.jpg,30
3075 | _xQKjTzQjxA_11258.jpg,79
3076 | 5K2dWgHU5wY_00150.jpg,79
3077 | e3XZzGyymyM_01092.jpg,38
3078 | 0s75lZ9gEnA_02190.jpg,79
3079 | NTHIGxzxMPg_03482.jpg,79
3080 | _VL19C7ClfA_01045.jpg,79
3081 | NTHIGxzxMPg_05649.jpg,79
3082 | uWVCi1XuPqA_00266.jpg,38
3083 | NTHIGxzxMPg_03607.jpg,79
3084 | F1O1IUhq4gs_08840.jpg,0
3085 | QjeVtQDPbqw_02872.jpg,79
3086 | NTHIGxzxMPg_03808.jpg,79
3087 | 2nATnbGQoqQ_02374.jpg,79
3088 | RWdRSURbEQM_00000.jpg,128
3089 | bD2xy2PlMBw_03627.jpg,2
3090 | NTHIGxzxMPg_03786.jpg,79
3091 | NTHIGxzxMPg_04088.jpg,79
3092 | T3uduNq1hpY_02878.jpg,79
3093 | NTHIGxzxMPg_01849.jpg,79
3094 | s-Fue-h1zlw_04095.jpg,79
3095 | bD2xy2PlMBw_02256.jpg,2
3096 | bD2xy2PlMBw_01800.jpg,23
3097 | 0s75lZ9gEnA_02880.jpg,79
3098 | QjeVtQDPbqw_09263.jpg,79
3099 | bD2xy2PlMBw_01567.jpg,0
3100 | YkKbl4SeD5I_00415.jpg,79
3101 | _xQKjTzQjxA_06833.jpg,79
3102 | bD2xy2PlMBw_03266.jpg,2
3103 | rQB4Hwi4O-M_00494.jpg,37
3104 | L-PHz9HWvmI_00000.jpg,17
3105 | rQB4Hwi4O-M_01443.jpg,37
3106 | pkVeJkLNRc8_05293.jpg,70
3107 | eQA0sfMhNis_04986.jpg,70
3108 | KDiC3ElmkDE_01147.jpg,70
3109 | 66NaQpdP2qI_03369.jpg,70
3110 | 66NaQpdP2qI_08230.jpg,70
3111 | eQA0sfMhNis_09652.jpg,6
3112 | eQA0sfMhNis_10176.jpg,23
3113 | KDiC3ElmkDE_03401.jpg,70
3114 | eQA0sfMhNis_04975.jpg,70
3115 | gdFztOz1zGg_00401.jpg,20
3116 | KDiC3ElmkDE_03379.jpg,70
3117 | KDiC3ElmkDE_06854.jpg,70
3118 | C4J0L7i7fGM_01213.jpg,7
3119 | KDiC3ElmkDE_06910.jpg,70
3120 | 0UV8E7vWxhU_00227.jpg,95
3121 | 4YspZcIRPXw_02253.jpg,86
3122 | 4YspZcIRPXw_00582.jpg,86
3123 | syxZ7Ncu7Rs_00073.jpg,86
3124 | QPklDWo-T1M_03997.jpg,86
3125 | QuRn0yzHRA8_01678.jpg,86
3126 | QuRn0yzHRA8_00984.jpg,86
3127 | _OgZFQcvg1Q_00696.jpg,135
3128 | 7tsBJpqoK_0_00496.jpg,127
3129 | E_BOTTHGCRo_00000.jpg,24
3130 | 7tsBJpqoK_0_04925.jpg,127
3131 | ejr3DnUUaQI_02675.jpg,127
3132 | HZjaLrUo1SQ_02652.jpg,127
3133 | 7tsBJpqoK_0_03950.jpg,127
3134 | HZjaLrUo1SQ_05388.jpg,24
3135 | 7tsBJpqoK_0_00548.jpg,127
3136 | 7tsBJpqoK_0_01357.jpg,127
3137 | eyHJ6kSrJ8o_00244.jpg,24
3138 | HZjaLrUo1SQ_04062.jpg,127
3139 | HZjaLrUo1SQ_05011.jpg,24
3140 | BtmXXvT7ESQ_01637.jpg,127
3141 | 7tsBJpqoK_0_00541.jpg,127
3142 | BtmXXvT7ESQ_00563.jpg,127
3143 | GLPbffLeKy0_00633.jpg,127
3144 | BtmXXvT7ESQ_00170.jpg,127
3145 | 7tsBJpqoK_0_04913.jpg,127
3146 | 7tsBJpqoK_0_01440.jpg,127
3147 | JkWoyZbRNj0_04447.jpg,170
3148 | JkWoyZbRNj0_04409.jpg,170
3149 | JkWoyZbRNj0_06836.jpg,170
3150 | LDnXtVyEMTY_01339.jpg,49
3151 | vva3cILHdFA_00000.jpg,19
3152 | 5H2X2My3EKc_00220.jpg,8
3153 | vSDVGHaJYWc_02738.jpg,98
3154 | C2aWqmIJlds_00798.jpg,49
3155 | 0aKXAnPy5s0_00435.jpg,119
3156 | UBc52w1k-dc_01833.jpg,119
3157 | NYFiglXtm7w_07868.jpg,119
3158 | PlRQGsx5qHs_01627.jpg,119
3159 | Z2s1aqxjBTY_02806.jpg,119
3160 | SoUU9MamDms_00360.jpg,0
3161 | 0aKXAnPy5s0_01557.jpg,10
3162 | VH6q6Jy-yzs_04079.jpg,119
3163 | rQRbDVG-ra0_00100.jpg,135
3164 | Xl6hK5OkwG8_02504.jpg,135
3165 | 3yzVVN43Z5o_02262.jpg,119
3166 | 3yzVVN43Z5o_01214.jpg,119
3167 | Z2s1aqxjBTY_01871.jpg,119
3168 | YjOq5g6oPUM_00313.jpg,119
3169 | EWIz4eYj3fE_00084.jpg,119
3170 | 3yzVVN43Z5o_01751.jpg,119
3171 | guMXsTsqDbk_00000.jpg,119
3172 | 5yGmRx3ymWI_00978.jpg,119
3173 | 95B8N7WGaoU_03590.jpg,205
3174 | djmf63oMiWs_05725.jpg,205
3175 | fIlot_yc8xQ_02023.jpg,0
3176 | 5zr-wFuvneE_05554.jpg,205
3177 | 95B8N7WGaoU_03626.jpg,205
3178 | 5zr-wFuvneE_10061.jpg,205
3179 | djmf63oMiWs_05730.jpg,205
3180 | e2SoivHIdXM_02955.jpg,38
3181 | e2SoivHIdXM_02378.jpg,197
3182 | dRQ_CmBTPsQ_00169.jpg,195
3183 | U4NceKwIhKQ_04972.jpg,195
3184 | 80F7RVezVhM_01115.jpg,195
3185 | dRQ_CmBTPsQ_02205.jpg,195
3186 | dRQ_CmBTPsQ_00000.jpg,195
3187 | RrDwhAJnegM_06221.jpg,195
3188 | dRQ_CmBTPsQ_00965.jpg,195
3189 | dRQ_CmBTPsQ_02079.jpg,195
3190 | TdEir3CYDMM_02871.jpg,128
3191 | PYUIHcnAgbc_02563.jpg,160
3192 | Vt9-JtK7II8_04289.jpg,160
3193 | PYUIHcnAgbc_07309.jpg,160
3194 | 0nkBJ5ZtZT8_01595.jpg,128
3195 | GEnUgOhYJgQ_03056.jpg,76
3196 | LHNPoSBJvn8_09036.jpg,160
3197 | Vt9-JtK7II8_04417.jpg,160
3198 | HEWBxdqhdmE_01745.jpg,160
3199 | PYUIHcnAgbc_08896.jpg,160
3200 | Jepm6rPRQPE_11937.jpg,160
3201 | 3aPpkPGzkMI_04338.jpg,160
3202 | 7C-qzaHmxWA_05731.jpg,160
3203 | Vt9-JtK7II8_03646.jpg,160
3204 | 8BNYmg9asQc_05124.jpg,160
3205 | PYzbif2-fGo_03568.jpg,51
3206 | NFHdej1LQrA_04932.jpg,79
3207 | 411PRE9wiXI_15727.jpg,51
3208 | 411PRE9wiXI_16363.jpg,51
3209 | 411PRE9wiXI_16079.jpg,51
3210 | St4OIQFJyFA_15727.jpg,51
3211 | CZp5IpT9mpk_00000.jpg,204
3212 | VPxAN5ud0O8_00943.jpg,110
3213 | VPxAN5ud0O8_01642.jpg,110
3214 | cR_wuVexbFg_06615.jpg,110
3215 | MQ98mvkKE4Q_06712.jpg,110
3216 | UUiEHkBOMd4_11826.jpg,39
3217 | UUiEHkBOMd4_08110.jpg,39
3218 | 1paf5vsB9a8_01613.jpg,39
3219 | UUiEHkBOMd4_11857.jpg,39
3220 | UUiEHkBOMd4_11581.jpg,39
3221 | UUiEHkBOMd4_11750.jpg,39
3222 | YcicN3xD-SU_01632.jpg,39
3223 | UUiEHkBOMd4_11833.jpg,39
3224 | 1paf5vsB9a8_01321.jpg,39
3225 | UUiEHkBOMd4_08394.jpg,39
3226 | UUiEHkBOMd4_11756.jpg,39
3227 | UUiEHkBOMd4_11767.jpg,39
3228 | UUiEHkBOMd4_04903.jpg,39
3229 | UUiEHkBOMd4_04958.jpg,39
3230 | UUiEHkBOMd4_08290.jpg,39
3231 | 7abf2Vyesrs_04212.jpg,11
3232 | UUiEHkBOMd4_11495.jpg,39
3233 | UUiEHkBOMd4_11664.jpg,39
3234 | UUiEHkBOMd4_04749.jpg,39
3235 | 17rZTpoXVzw_01473.jpg,39
3236 | UUiEHkBOMd4_11690.jpg,39
3237 | YcicN3xD-SU_04185.jpg,39
3238 | TAOrHz9s1gQ_04956.jpg,39
3239 | UUiEHkBOMd4_11482.jpg,39
3240 | rZWpnx6UKBs_00436.jpg,39
3241 | 17rZTpoXVzw_00112.jpg,39
3242 | 17rZTpoXVzw_01083.jpg,39
3243 | LOTQJVWxHSI_00651.jpg,39
3244 | UUiEHkBOMd4_07973.jpg,39
3245 | UUiEHkBOMd4_11806.jpg,39
3246 | TAOrHz9s1gQ_12367.jpg,39
3247 | QV-36hZraoY_01385.jpg,39
3248 | 0VXjHETo-20_00146.jpg,39
3249 | UUiEHkBOMd4_11763.jpg,39
3250 | c4Hx_y1-3PU_05486.jpg,157
3251 | c4Hx_y1-3PU_05479.jpg,157
3252 | 1AQ9cZGVDGM_01944.jpg,157
3253 | QLh5aaM6wi8_01308.jpg,157
3254 | Z9q77Oa5iJs_00349.jpg,157
3255 | ZMlK9lunqE0_02415.jpg,166
3256 | ZlfZWhzAfn4_00282.jpg,128
3257 | hvFIlGmuIgE_01430.jpg,166
3258 | GAP-SpfKebg_01181.jpg,166
3259 | 0nifAVL-CmQ_00390.jpg,150
3260 | il-u8o8ebl8_02773.jpg,166
3261 | ZMlK9lunqE0_02473.jpg,166
3262 | hvFIlGmuIgE_01143.jpg,166
3263 | BsNZwmq1AXU_00673.jpg,172
3264 | LWRMOJQDiLU_11255.jpg,4
3265 | JOx8qn54nZs_04850.jpg,172
3266 | QUqhoN-V58k_02525.jpg,57
3267 | BsNZwmq1AXU_05135.jpg,172
3268 | 18JHobzUDzU_02809.jpg,166
3269 | q5V6xoFAQZc_01017.jpg,166
3270 | GpajsZjLu-U_05413.jpg,45
3271 | 9nTlgtf7TME_05244.jpg,62
3272 | 9nTlgtf7TME_04747.jpg,62
3273 | Apq9RHrwfWs_00647.jpg,95
3274 | tiNl99Z01MQ_02338.jpg,2
3275 | cRwFTtOWAOM_06178.jpg,162
3276 | cRwFTtOWAOM_08845.jpg,162
3277 | cRwFTtOWAOM_07351.jpg,162
3278 | HE3SZBxVNlI_02584.jpg,31
3279 | HE3SZBxVNlI_02624.jpg,31
3280 | gA0bJLxWIB0_04772.jpg,128
3281 | LRNxKtmlFaI_01617.jpg,203
3282 | 2ThXvnlhV3c_00240.jpg,92
3283 | 3BiYQbD-mVk_03706.jpg,98
3284 | ogMVEakZUFQ_00841.jpg,27
3285 | fFLwOWe0KQ4_08599.jpg,18
3286 | 87uKpqfVe78_03171.jpg,94
3287 | 87uKpqfVe78_03318.jpg,94
3288 | LyIn9d6c880_06086.jpg,18
3289 | fFLwOWe0KQ4_11873.jpg,18
3290 | 5X6e7Su9CdQ_00976.jpg,0
3291 | Jz2N90cO0Eo_00072.jpg,18
3292 | HiyTlTrxqQA_00284.jpg,140
3293 | xxJtR5En9ps_00471.jpg,7
3294 | qEgUJ8owOss_01044.jpg,18
3295 | HiyTlTrxqQA_05459.jpg,140
3296 | nh2HSuORymg_00885.jpg,18
3297 | xxJtR5En9ps_00516.jpg,18
3298 | 5X6e7Su9CdQ_05905.jpg,18
3299 | bgr9p7wQRek_01491.jpg,8
3300 | ZoUrM_ZXoOE_00087.jpg,8
3301 | LyIn9d6c880_06832.jpg,18
3302 | Jz2N90cO0Eo_01057.jpg,18
3303 | ynUNOsMrO1k_00908.jpg,7
3304 | 3b2YX4OnR2k_00859.jpg,18
3305 | Jz2N90cO0Eo_01667.jpg,18
3306 | xxJtR5En9ps_00859.jpg,18
3307 | xxJtR5En9ps_01270.jpg,18
3308 | fFLwOWe0KQ4_00000.jpg,18
3309 | fFLwOWe0KQ4_12263.jpg,18
3310 | RHw-r3z-1p4_00000.jpg,2
3311 | dI3Lcfl4Fko_00387.jpg,2
3312 | V-mtbJXjSno_03392.jpg,135
3313 | SO1VZntbkYc_00335.jpg,11
3314 | Us_TXFDaVJo_03104.jpg,35
3315 | 8SWBXH7xvyY_00564.jpg,35
3316 | hB8hTar1Mv0_00954.jpg,128
3317 | cbUjhDA9luI_00848.jpg,8
3318 | SQ7eaUJtEIs_01260.jpg,144
3319 | ebYt-jlCeL4_01574.jpg,22
3320 | pgPgey7EudE_02291.jpg,144
3321 | 1L1dYO8EG74_07102.jpg,164
3322 | 5Dle76vhHNM_05568.jpg,171
3323 | 1L1dYO8EG74_05375.jpg,164
3324 | IIeP7dNkoes_01427.jpg,173
3325 | iVmSweQwzgs_01810.jpg,120
3326 | iTnJWiuh9eU_00000.jpg,128
3327 | duEViRHxiNk_07772.jpg,128
3328 | bNGX8hRaChk_05168.jpg,18
3329 | z7Ebe9A2oOg_01590.jpg,65
3330 | 9uMtkrrNgzQ_04269.jpg,0
3331 | 0-hzMzEaVc8_00730.jpg,89
3332 | HHMoAfNJH3Y_04290.jpg,9
3333 | NVIgqAbZJJc_10166.jpg,89
3334 | EpHM_a0NKQU_11094.jpg,1
3335 | HaheCj5DKFg_01949.jpg,116
3336 | 3EmbtOXT8Ks_01512.jpg,116
3337 | 3EmbtOXT8Ks_00058.jpg,4
3338 | 4rZvvkqnatc_02974.jpg,2
3339 | GDzsuBgQxY0_01080.jpg,2
3340 | qeH5wJ3L7Ts_02828.jpg,128
3341 | HADHCxEbQPs_03587.jpg,16
3342 | 2yaMc9Sz-YE_04019.jpg,16
3343 | UY8svVboV34_00000.jpg,204
3344 | HADHCxEbQPs_00644.jpg,16
3345 | HADHCxEbQPs_02680.jpg,16
3346 | Mqn5R5rrwdI_00239.jpg,177
3347 | Mqn5R5rrwdI_00000.jpg,177
3348 | UUD5aq4dBf8_00262.jpg,94
3349 | a-jzCDIryfQ_04400.jpg,115
3350 | 8w5uGcPAIl0_05690.jpg,103
3351 | BFIwtMjhEiY_00069.jpg,6
3352 | ec-xlDTTjOs_02185.jpg,60
3353 | ec-xlDTTjOs_02538.jpg,60
3354 | pvMQOW44vMU_00396.jpg,128
3355 | H5ma41Abra8_04538.jpg,158
3356 | 4ckVVmPvuZI_09378.jpg,8
3357 | 1TYstYNCpEI_15384.jpg,22
3358 | epwvJAICClA_02201.jpg,120
3359 | fibYErY9uX0_00058.jpg,1
3360 | 9-xTeA6kFa8_00823.jpg,7
3361 | 5EmrvBV3P94_01056.jpg,96
3362 | _DjVFex5NWY_02334.jpg,8
3363 | xjD8yqoFEmQ_00000.jpg,6
3364 | 5jNBtQjvNgg_02080.jpg,203
3365 | 5jNBtQjvNgg_00878.jpg,203
3366 | 5jNBtQjvNgg_00561.jpg,203
3367 | VRxjKEjNvIo_01513.jpg,27
3368 | VRxjKEjNvIo_02935.jpg,27
3369 | E_sxDFQc3s0_06834.jpg,27
3370 | kSmNqfMEL84_00377.jpg,27
3371 | VRxjKEjNvIo_02262.jpg,27
3372 | IVMM7VJwN7o_156000.jpg,30
3373 | TnjxURJq6P8_00259.jpg,27
3374 | VRxjKEjNvIo_02321.jpg,27
3375 | TnjxURJq6P8_01464.jpg,27
3376 | lyh0kFcZaZE_09136.jpg,106
3377 | lyh0kFcZaZE_10110.jpg,106
3378 | lyh0kFcZaZE_06204.jpg,106
3379 | 5odjHyOPm7c_00173.jpg,32
3380 | _aEyGpXyDjg_00302.jpg,38
3381 | 5odjHyOPm7c_00125.jpg,32
3382 | yxq6pYK-fGg_00438.jpg,78
3383 | wQ63UH4Kp9k_02919.jpg,175
3384 | AqX4Eb33jG8_03789.jpg,200
3385 | _NGLacFqKf0_03612.jpg,200
3386 | _NGLacFqKf0_01796.jpg,200
3387 | AqX4Eb33jG8_05054.jpg,200
3388 | Thpjk69h9P8_00577.jpg,17
3389 | Z68Rulp1fR0_01430.jpg,169
3390 | rugOj1duAeE_02809.jpg,128
3391 | f1SlbCxQjxY_03377.jpg,11
3392 | rugOj1duAeE_00203.jpg,81
3393 | rugOj1duAeE_03021.jpg,81
3394 | rugOj1duAeE_04209.jpg,81
3395 | 9yzxjzPB5F0_01233.jpg,197
3396 | 9yzxjzPB5F0_01469.jpg,197
3397 | DfhwmcVqCP8_00348.jpg,197
3398 | b-MbEx1Iqj8_01052.jpg,111
3399 | r7Gwaoru_gg_01089.jpg,111
3400 | AlvGmMZCUMQ_00499.jpg,180
3401 | AlvGmMZCUMQ_04892.jpg,180
3402 | 


--------------------------------------------------------------------------------
/run.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | TIME=$(date "+%Y-%m-%d-%H-%M-%S")
 3 | 
 4 | 
 5 | OUTPUT_PATH=./outputs
 6 | TRAIN_LIST=/home/lxztju/pytorch_classification/sample_files/imgs/listfile.txt
 7 | VAL_LIST=/home/lxztju/pytorch_classification/sample_files/imgs/listfile.txt
 8 | 
 9 | #train
10 | # CUDA_VISIBLE_DEVICES=0,1,2,3 \
11 | # python3 -u -m torch.distributed.launch --nproc_per_node 4  ./tools/train_val.py \
12 | #     --model_name=resnet18 \
13 | #     --lr  0.01 --epochs 70  --batch-size 128  -j 4 \
14 | #     --output=$OUTPUT_PATH/$TIME \
15 | #     --train_list=$TRAIN_LIST \
16 | #     --val_list=$VAL_LIST \
17 | #     --num_classes=2 \
18 | #     --is_pretrained
19 |     
20 |     
21 |     
22 |     
23 | # CUDA_VISIBLE_DEVICES=0,1,2,3 \
24 | # python3 -u -m torch.distributed.launch --nproc_per_node 1  ./tools/evaluation.py \
25 | #     --model_name=resnet18 \
26 | #     --batch-size 64  -j 4 \
27 | #     --output=$OUTPUT_PATH/$TIME \
28 | #     --val_list=$VAL_LIST \
29 | #     --tune_from='/home/lxztju/pytorch_classification/ouputs/xxx/epoch_4.pth' \
30 | #     --num_classes=2
31 |     
32 | 
33 | # CUDA_VISIBLE_DEVICES=0,1,2,3 \
34 | # python3 -u -m torch.distributed.launch --nproc_per_node 1  ./tools/predict.py \
35 | #     --model_name=resnet18 \
36 | #     --batch-size 64  -j 4 \
37 | #     --output=$OUTPUT_PATH/$TIME \
38 | #     --val_list=$VAL_LIST \
39 | #     --tune_from='/home/lxztju/pytorch_classification/ouputs/xxx/epoch_4.pth' \
40 | #     --num_classes=2
41 | 


--------------------------------------------------------------------------------
/sample_files/imgs/cat/0.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/cat/0.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/cat/1.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/cat/1.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/cat/2.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/cat/2.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/cat/3.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/cat/3.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/cat/4.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/cat/4.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/cat/5.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/cat/5.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/dog/0.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/dog/0.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/dog/1.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/dog/1.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/dog/2.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/dog/2.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/dog/3.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/dog/3.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/dog/4.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/dog/4.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/dog/5.jpg:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/sample_files/imgs/dog/5.jpg


--------------------------------------------------------------------------------
/sample_files/imgs/listfile.txt:
--------------------------------------------------------------------------------
 1 | /home/lxztju/pytorch_classification/sample_files/imgs/cat/0.jpg	0
 2 | /home/lxztju/pytorch_classification/sample_files/imgs/cat/1.jpg	0
 3 | /home/lxztju/pytorch_classification/sample_files/imgs/cat/2.jpg	0
 4 | /home/lxztju/pytorch_classification/sample_files/imgs/cat/3.jpg	0
 5 | /home/lxztju/pytorch_classification/sample_files/imgs/cat/4.jpg	0
 6 | /home/lxztju/pytorch_classification/sample_files/imgs/cat/5.jpg	0
 7 | /home/lxztju/pytorch_classification/sample_files/imgs/dog/0.jpg	1
 8 | /home/lxztju/pytorch_classification/sample_files/imgs/dog/1.jpg	1
 9 | /home/lxztju/pytorch_classification/sample_files/imgs/dog/2.jpg	1
10 | /home/lxztju/pytorch_classification/sample_files/imgs/dog/3.jpg	1
11 | /home/lxztju/pytorch_classification/sample_files/imgs/dog/4.jpg	1
12 | /home/lxztju/pytorch_classification/sample_files/imgs/dog/5.jpg	1
13 | 


--------------------------------------------------------------------------------
/tools/evaluation.py:
--------------------------------------------------------------------------------
  1 | import os
  2 | import tqdm
  3 | import time
  4 | import shutil
  5 | 
  6 | 
  7 | import numpy as np
  8 | import torch.nn.parallel
  9 | import torch.backends.cudnn as cudnn
 10 | import torch.optim
 11 | from torch.nn import functional as F
 12 | import torch.distributed as dist
 13 | 
 14 | import sys
 15 | from pathlib import Path
 16 | FILE = Path(__file__).resolve()
 17 | 
 18 | ROOT = FILE.parents[1]
 19 | if str(ROOT) not in sys.path:
 20 |     sys.path.append(str(ROOT))  # add ROOT to PATH
 21 | 
 22 | from utils import  init_logger, torch_distributed_zero_first, AverageMeter, distributed_concat
 23 | from utils import  get_scheduler, parser
 24 |         
 25 | from dataset import ClsDataset, train_transform, val_transform
 26 | from cls_models import ClsModel
 27 | 
 28 | def evaluate(rank, local_rank, device, args):
 29 |     
 30 |     check_rootfolders()
 31 |     logger = init_logger(log_file=args.output + f'/log', rank=rank)
 32 |     
 33 |     with torch_distributed_zero_first(rank):
 34 |         val_dataset = ClsDataset(
 35 |             list_file = args.val_list,
 36 |             transform = val_transform(size=args.input_size)
 37 |         )
 38 | 
 39 |     val_sampler = torch.utils.data.distributed.DistributedSampler(val_dataset, rank=rank,shuffle=False)
 40 | 
 41 |     val_loader = torch.utils.data.DataLoader(
 42 |         val_dataset,
 43 |         batch_size=args.batch_size,
 44 |         sampler=val_sampler,
 45 |         num_workers=args.workers, pin_memory=True)
 46 | 
 47 |     print('val_loader is ready!!!')
 48 | 
 49 |         
 50 |     model = ClsModel(args.model_name, args.num_classes, args.is_pretrained)
 51 | 
 52 | 
 53 |     if args.tune_from and os.path.exists(args.tune_from):
 54 |         print(f'loading model from {args.tune_from}')
 55 |         sd = torch.load(args.tune_from, map_location='cpu')
 56 |         model.load_state_dict(sd)
 57 |     else:
 58 |         raise ValueError("the path of model weights is not exist!")
 59 |         
 60 |     model.to(device)
 61 |     model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_rank], output_device=local_rank, find_unused_parameters=True)     
 62 |     
 63 |     cudnn.benchmark = True
 64 | 
 65 |         
 66 |     for k, v in sorted(vars(args).items()):
 67 |         logger.info(f'{k} = {v}')
 68 | 
 69 |     model.eval()
 70 |     with torch.no_grad():
 71 |         preds, labels, scores = [], [], []
 72 |         eval_pbar = tqdm.tqdm(val_loader, desc=f'evaluating', position=1, disable=False if rank in [-1, 0] else True)
 73 |         for step, (img, target, _) in enumerate(eval_pbar):
 74 |             img = img.to(device)
 75 |             target = target.to(device)
 76 | 
 77 |             output = model(img)
 78 | 
 79 |             score = torch.softmax(output, dim=1)
 80 |             predict = torch.max(output, dim=1)[1]
 81 |             labels.append(target)
 82 |             scores.append(score)
 83 |             preds.append(predict)
 84 |         labels = torch.cat(labels, dim=0)
 85 |         predicts = torch.cat(preds, dim=0)
 86 |         scores = torch.cat(scores, dim=0)
 87 |         if rank != -1:
 88 |             labels = distributed_concat(labels, len(val_dataset))
 89 |             predicts = distributed_concat(predicts, len(val_dataset))
 90 |             scores = distributed_concat(scores, len(val_dataset))
 91 |         scores = scores.cpu().numpy()
 92 |         labels = labels.cpu().numpy()
 93 |         predicts = predicts.cpu().numpy()
 94 |         if rank == 0:
 95 |             from sklearn import metrics
 96 |             report = metrics.classification_report(labels, predicts, target_names=['{}'.format(x) for x in range(args.num_classes)],
 97 |                                                    digits=4, labels=range(args.num_classes))
 98 |             
 99 |             confusion = metrics.confusion_matrix(labels, predicts)
100 |             print(report)
101 |             print(confusion)
102 |             performance = np.sum(labels==predicts) / len(labels)
103 |             print(performance)
104 |             np.save(os.path.join(args.output, f"labels"), labels)
105 |             np.save(os.path.join(args.output, f"scores"), scores)
106 |             np.save(os.path.join(args.output, f"predicts"), predicts)
107 |             logger.info(args.output)
108 | 
109 |             
110 | def check_rootfolders():
111 |     """Create log and model folder"""
112 |     folders_util = [args.output]
113 |     for folder in folders_util:
114 |         os.makedirs(folder, exist_ok=True)
115 | 
116 | 
117 | def distributed_init(backend="gloo", port=None):
118 | 
119 |     num_gpus = torch.cuda.device_count()
120 | 
121 |     rank = int(os.environ["RANK"])
122 |     world_size = int(os.environ["WORLD_SIZE"])
123 | 
124 |     torch.cuda.set_device(rank % num_gpus)
125 | 
126 |     dist.init_process_group(
127 |         backend=backend,
128 |         world_size=world_size,
129 |         rank=rank,
130 |     )
131 |             
132 | if __name__ == '__main__':
133 |     
134 |     args = parser.parse_args()
135 |     distributed_init(backend = args.backend)
136 |     rank = int(os.environ["RANK"])
137 |     local_rank = int(os.environ["LOCAL_RANK"])
138 |     device = torch.device("cuda", local_rank)
139 | 
140 |     args.world_size = int(os.environ["WORLD_SIZE"])
141 |     
142 |     print(f"[init] == local rank: {local_rank}, global rank: {rank} == devices: {device}")
143 |     
144 |     evaluate(rank, local_rank, device, args)


--------------------------------------------------------------------------------
/tools/predict.py:
--------------------------------------------------------------------------------
 1 | import os
 2 | import tqdm
 3 | import time
 4 | import shutil
 5 | 
 6 | 
 7 | import numpy as np
 8 | import torch.nn.parallel
 9 | import torch.backends.cudnn as cudnn
10 | import torch.optim
11 | from torch.nn import functional as F
12 | import torch.distributed as dist
13 | from PIL import Image
14 | 
15 | import sys
16 | from pathlib import Path
17 | FILE = Path(__file__).resolve()
18 | 
19 | ROOT = FILE.parents[1]
20 | if str(ROOT) not in sys.path:
21 |     sys.path.append(str(ROOT))  # add ROOT to PATH
22 | 
23 | from utils import  init_logger, torch_distributed_zero_first, AverageMeter, distributed_concat
24 | from utils import  get_scheduler, parser
25 |         
26 | from dataset import ClsDataset, train_transform, val_transform
27 | from cls_models import ClsModel
28 | 
29 | 
30 | def __init_model(args):
31 |     logger = init_logger(log_file=args.output + f'/log', rank=-1)
32 | 
33 |     model = ClsModel(args.model_name, args.num_classes)
34 |     if args.tune_from and os.path.exists(args.tune_from):
35 |         print(f'loading model from {args.tune_from}')
36 |         sd = torch.load(args.tune_from, map_location='cpu')
37 |         model.load_state_dict(sd)
38 |     device = torch.device("cuda")
39 |     model.to(device)
40 |     model.eval()
41 | 
42 |     cudnn.benchmark = True
43 |     return logger, model
44 | 
45 | 
46 | 
47 | def predict(img_path):
48 |     
49 |     img = Image.open(img_path).convert('RGB')
50 |     img_tensor = val_transform(size=args.input_size)(img).unsqueeze(0)
51 | 
52 |     
53 |     with torch.no_grad():
54 |         preds, labels, scores = [], [], []
55 |         device = torch.device("cuda")
56 |         img_tensor = img_tensor.to(device)
57 | 
58 |         output = model(img_tensor)
59 | 
60 |         scores = torch.softmax(output, dim=1)
61 |         score = torch.max(scores, dim=1)[0].item()
62 |         pred = torch.max(scores, dim=1)[1].item()
63 |         
64 |     return pred, score
65 |         
66 |         
67 | def check_rootfolders():
68 |     """Create log and model folder"""
69 |     folders_util = [args.output]
70 |     for folder in folders_util:
71 |         os.makedirs(folder, exist_ok=True)
72 | 
73 |             
74 |             
75 | if __name__ == '__main__':
76 |     
77 |     args = parser.parse_args()
78 |     check_rootfolders()
79 |     logger, model = __init_model(args)
80 |     
81 |     for k, v in sorted(vars(args).items()):
82 |         logger.info(f'{k} = {v}')
83 |         
84 |     pred_file = args.val_list
85 |     
86 |     datas = open(pred_file, 'r').readlines()
87 |     target_res = open(os.path.join(args.output, 'pred_res.txt'), 'w')
88 |     for data in tqdm.tqdm(datas):
89 |         path, _ = data.strip().split('\t')
90 |         pred, score = predict(path)
91 | #         print(pred, score)
92 |         target_res.write(path + '\t' + str(pred) + '\t' + str(score) +'\n')
93 |         target_res.flush()
94 |     target_res.close()


--------------------------------------------------------------------------------
/tools/train_val.py:
--------------------------------------------------------------------------------
  1 | import os
  2 | import tqdm
  3 | import time
  4 | import shutil
  5 | 
  6 | 
  7 | import numpy as np
  8 | import torch.nn.parallel
  9 | import torch.backends.cudnn as cudnn
 10 | import torch.optim
 11 | from torch.nn import functional as F
 12 | import torch.distributed as dist
 13 | 
 14 | import sys
 15 | from pathlib import Path
 16 | FILE = Path(__file__).resolve()
 17 | 
 18 | ROOT = FILE.parents[1]
 19 | if str(ROOT) not in sys.path:
 20 |     sys.path.append(str(ROOT))  # add ROOT to PATH
 21 | 
 22 | from utils import  init_logger, torch_distributed_zero_first, AverageMeter, distributed_concat
 23 | from utils import  get_scheduler, parser
 24 | 
 25 | from dataset import ClsDataset, train_transform, val_transform
 26 | from cls_models import ClsModel
 27 | 
 28 | 
 29 | def train(rank, local_rank, device, args):
 30 |     check_rootfolders()
 31 |     logger = init_logger(log_file=args.output + f'/log', rank=rank)
 32 | 
 33 |     with torch_distributed_zero_first(rank):
 34 |         val_dataset = ClsDataset(
 35 |             list_file = args.val_list,
 36 |             transform = val_transform(size=args.input_size)
 37 |         )
 38 | 
 39 |         train_dataset = ClsDataset(
 40 |             list_file = args.train_list,
 41 |             transform = train_transform(size=args.input_size)
 42 |         )
 43 | 
 44 |         logger.info(f"Num train examples = {len(train_dataset)}")
 45 |         logger.info(f"Num val examples = {len(val_dataset)}")
 46 | 
 47 | 
 48 |     val_sampler = torch.utils.data.distributed.DistributedSampler(val_dataset, rank=rank,shuffle=False)
 49 | 
 50 |     val_loader = torch.utils.data.DataLoader(
 51 |         val_dataset,
 52 |         batch_size=args.batch_size,
 53 |         sampler=val_sampler,
 54 |         num_workers=args.workers, pin_memory=True)
 55 | 
 56 | 
 57 |     train_sampler = torch.utils.data.distributed.DistributedSampler(train_dataset, rank=rank,shuffle=True)
 58 |     train_loader = torch.utils.data.DataLoader(train_dataset,
 59 |         batch_size=args.batch_size,
 60 |         sampler=train_sampler,
 61 |         num_workers=args.workers, pin_memory=True,
 62 |         drop_last=True)
 63 | 
 64 | 
 65 |     criterion = torch.nn.CrossEntropyLoss().to(device)
 66 | 
 67 |     model = ClsModel(args.model_name, args.num_classes, args.is_pretrained)
 68 |     print(model.base_model)
 69 |     model.to(device)
 70 |     model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[local_rank], output_device=local_rank, find_unused_parameters=True)
 71 | 
 72 |     optimizer = torch.optim.SGD(model.parameters(),lr=args.lr,momentum=args.momentum,weight_decay=args.weight_decay)
 73 |     scheduler = get_scheduler(optimizer,len(train_loader), args)
 74 | 
 75 | 
 76 |     cudnn.benchmark = True
 77 | 
 78 |     for k, v in sorted(vars(args).items()):
 79 |         logger.info(f'{k} = {v}')
 80 | 
 81 |     epoch = args.start_epoch
 82 |     model.zero_grad()
 83 |     eval_results = []
 84 |     for epoch in range(args.start_epoch, args.epochs):
 85 |         losses = AverageMeter()
 86 |         if local_rank != -1:
 87 |             train_sampler.set_epoch(epoch)
 88 |         model.train()
 89 |         for step, (img, target, _) in enumerate(train_loader):
 90 |             img = img.to(device)
 91 |             target = target.to(device)
 92 | 
 93 |             output  = model(img)
 94 |             loss = criterion(output, target)
 95 |             loss.backward()
 96 |             losses.update(loss.item(), img.size(0))
 97 |             if rank == 0:
 98 |                 if step % args.print_freq == 0:
 99 |                     logger.info(f"Epoch: [{epoch}/{args.epochs}][{step}/{len(train_loader)}], lr: {optimizer.param_groups[-1]['lr']:.5f} \t loss = {losses.val:.4f}({losses.avg:.4f})" )
100 |             optimizer.step()
101 |             optimizer.zero_grad()
102 |             scheduler.step()
103 | 
104 | 
105 |         # evaluate on validation set
106 |         if (epoch + 1) % args.eval_freq == 0 or epoch == args.epochs - 1:
107 |             model.eval()
108 |             with torch.no_grad():
109 |                 preds, labels = [], []
110 |                 eval_pbar = tqdm.tqdm(val_loader, desc=f'epoch {epoch + 1} / {args.epochs} evaluating', position=1, disable=False if rank in [-1, 0] else True)
111 |                 for step, (img, target, _) in enumerate(eval_pbar):
112 |                     img = img.to(device)
113 |                     target = target.to(device)
114 | 
115 |                     output = model(img)
116 |                     predict = torch.max(output, dim=1)[1]
117 | 
118 |                     labels.append(target)
119 |                     preds.append(predict)
120 | 
121 |                 labels = torch.cat(labels, dim=0)
122 |                 predicts = torch.cat(preds, dim=0)
123 |                 if rank != -1:
124 |                     labels = distributed_concat(labels, len(val_dataset))
125 |                     predicts = distributed_concat(predicts, len(val_dataset))
126 | 
127 | 
128 |                 labels = labels.cpu().numpy()
129 |                 predicts = predicts.cpu().numpy()
130 | 
131 |                 if rank == 0:
132 |                     eval_result = (np.sum(labels == predicts)) / len(labels)
133 |                     eval_results.append(eval_result)
134 |                     logger.info(f'precision = {eval_result:.4f}' )
135 |                     save_path = os.path.join(args.output, f'precision_{eval_result:.4f}_num_{epoch+1}')
136 |                     os.makedirs(save_path, exist_ok=True)
137 |                     model_to_save = (model.module if hasattr(model, "module") else model)
138 |                     torch.save(model_to_save.state_dict(), os.path.join(save_path, f'epoch_{epoch+1}.pth'))
139 | 
140 | 
141 | def check_rootfolders():
142 |     """Create log and model folder"""
143 |     folders_util = [args.output]
144 |     for folder in folders_util:
145 |         os.makedirs(folder, exist_ok=True)
146 | 
147 | 
148 | def distributed_init(backend="gloo", port=None):
149 | 
150 |     num_gpus = torch.cuda.device_count()
151 | 
152 |     rank = int(os.environ["RANK"])
153 |     world_size = int(os.environ["WORLD_SIZE"])
154 | 
155 |     torch.cuda.set_device(rank % num_gpus)
156 | 
157 |     dist.init_process_group(
158 |         backend=backend,
159 |         world_size=world_size,
160 |         rank=rank,
161 |     )
162 | 
163 | 
164 | if __name__ == '__main__':
165 | 
166 |     args = parser.parse_args()
167 |     distributed_init(backend = args.backend)
168 |     rank = int(os.environ["RANK"])
169 |     local_rank = int(os.environ["LOCAL_RANK"])
170 |     device = torch.device("cuda", local_rank)
171 |     print(args.train_list)
172 |     args.world_size = int(os.environ["WORLD_SIZE"])
173 | 
174 |     print(f"[init] == local rank: {local_rank}, global rank: {rank} == devices: {device}")
175 | 
176 |     train(rank, local_rank, device, args)
177 | 


--------------------------------------------------------------------------------
/trt_inference/CMakeLists.txt:
--------------------------------------------------------------------------------
 1 | cmake_minimum_required(VERSION 3.5)
 2 | project(hello_trt)
 3 | 
 4 | set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/lib)
 5 | set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/lib)
 6 | set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/bin)
 7 | set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
 8 | SET(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};-gencode arch=compute_61,code=sm_61;)
 9 | 
10 | 
11 | message(STATUS "CMAKE_RUNTIME_OUTPUT_DIRECTORY: "${CMAKE_RUNTIME_OUTPUT_DIRECTORY})
12 | 
13 | set(3RDPARTY ${CMAKE_CURRENT_SOURCE_DIR}/3rd)
14 | set(TENSORRT_LIB ${3RDPARTY}/TensorRT-8.0.1.6/targets/x86_64-linux-gnu/lib)
15 | file(GLOB TENSORRT_LIBS "${TENSORRT_LIB}/*.so")
16 | 
17 | set(OpenCV_DIR ${3RDPARTY}/opencv-3.4.16/share/OpenCV)
18 | find_package(CUDA REQUIRED)
19 | find_package(OpenCV REQUIRED)
20 | 
21 | message(STATUS "CUDA_LIBRARIES: "${CUDA_LIBRARIES})
22 | message(STATUS "TENSORRT_LIBS: "${TENSORRT_LIBS})
23 | message(STATUS "Opencv_LIBS: "${OpenCV_LIBS})
24 | include_directories(
25 | 	${CUDA_INCLUDE_DIRS}
26 |     ${3RDPARTY}/TensorRT-8.0.1.6/include
27 | 	${CMAKE_CURRENT_SOURCE_DIR}/include
28 |     ${OpenCV_INCLUDE_DIRS}
29 | 	)
30 | 
31 | add_executable(
32 |     trt_cls
33 |     src/TrtCls.cpp
34 |     )
35 | 
36 | target_link_libraries(
37 |    trt_cls
38 |     ${CUDA_LIBRARIES}
39 |     ${TENSORRT_LIBS}
40 |     ${OpenCV_LIBS}
41 |     )
42 | 


--------------------------------------------------------------------------------
/trt_inference/convert_onnx.py:
--------------------------------------------------------------------------------
 1 | import torch
 2 | import torch.onnx
 3 | 
 4 | 
 5 | def traced_model_convert_onnx():
 6 |     model = torch.jit.load('../cpp_inference/traced_model/traced_model_res50.pt')
 7 | 
 8 |     dummy_input = torch.randn(1, 3, 224, 224)
 9 | 
10 |     torch.onnx.export(
11 |         model,  # model
12 |         dummy_input, # 模型输入
13 |         "./saved_model/traced_res50.onnx", # 转换后模型的存储路径
14 |         export_params=True, # 是否将模型参数保存至onnx中
15 |         input_names=["input_image"], # 输入节点名称
16 |         output_names = ["model_output"], # 输出节点名称
17 |         dynamic_axes = {
18 |             "input_image":{0: "batch"}, # 可动态shape的维度
19 |             "model_output":{0: "batch"}
20 |                         },
21 |         opset_version=11 # onnx的版本
22 |     )
23 | 
24 | 
25 | import sys
26 | sys.path.append('/home/netease/codes/pytorch_classification')
27 | print(sys.path)
28 | from cls_models import ClsModel
29 | 
30 | def torch_convert_onnx():
31 | 
32 |     model = ClsModel('resnet50', num_classes=2, dropout=0, is_pretrained=False)
33 |     sd = torch.load('../cpp_inference/traced_model/trained_model.pth', map_location='cpu')
34 |     model.load_state_dict(sd)
35 |     model.eval()
36 | 
37 |     dummy_input = torch.randn(1, 3, 224, 224)
38 |     torch.onnx.export(
39 |         model,
40 |         dummy_input,
41 |         "./saved_model/torch_res50.onnx",
42 |         export_params=True,
43 |         input_names=["input_image"],
44 |         output_names = ["model_output"],
45 |         dynamic_axes = {
46 |             "input_image":{0: "batch"},
47 |             "model_output":{0: "batch"}
48 |                         },
49 |         opset_version=11
50 |     )
51 | 
52 | if __name__ == '__main__':
53 |     import os
54 |     os.makedirs('./saved_model', exist_ok=True)
55 |     torch_convert_onnx()
56 |     traced_model_convert_onnx()
57 | 
58 | 
59 | 


--------------------------------------------------------------------------------
/trt_inference/include/TrtCls.hpp:
--------------------------------------------------------------------------------
  1 | #ifndef TRTCLS
  2 | #define TRTCLS
  3 | 
  4 | #include <string>
  5 | #include <cstdio>
  6 | #include <memory>
  7 | #include <fstream>
  8 | #include <vector>
  9 | #include <numeric>
 10 | #include <cmath>
 11 | #include <sstream>
 12 | #include <unordered_map>
 13 | 
 14 | #include <opencv2/opencv.hpp>
 15 | #include <cuda_runtime.h>
 16 | #include <NvInfer.h>  
 17 | //#include <NvInferRuntime.h> NvInfer.h头文件中包含了这个头文件
 18 | 
 19 | 
 20 | 
 21 | inline const char* severity_string(nvinfer1::ILogger::Severity t){
 22 |     switch(t){
 23 |         case nvinfer1::ILogger::Severity::kINTERNAL_ERROR: return "internal_error";
 24 |         case nvinfer1::ILogger::Severity::kERROR: return "error";
 25 |         case nvinfer1::ILogger::Severity::kWARNING: return "warning";
 26 |         case nvinfer1::ILogger::Severity::kINFO: return "info";
 27 |         case nvinfer1::ILogger::Severity::kVERBOSE: return "verbose";
 28 |     }
 29 | }
 30 | 
 31 | class TrtLogger: public nvinfer1::ILogger{
 32 | public:
 33 |     virtual void log(Severity severity, nvinfer1::AsciiChar const* msg) noexcept override{
 34 |         if (severity <= Severity::kINFO){
 35 |             printf("%s : %s \n", severity_string(severity), msg);
 36 |         }
 37 |     }
 38 | };
 39 | 
 40 | // enum	Severity : int32_t {
 41 |   // Severity::kINTERNAL_ERROR = 0, 
 42 |   // Severity::kERROR = 1, 
 43 |   // Severity::kWARNING = 2, 
 44 |   // Severity::kINFO = 3,
 45 |   // Severity::kVERBOSE = 4
 46 | // }
 47 | 
 48 | class TrtCls{
 49 | public:
 50 |     TrtCls();
 51 |     ~TrtCls();
 52 | 
 53 | public:
 54 |     int build_engine(const std::string onnx_path, const std::string trt_engine_path);
 55 | 
 56 |     int init_engine(const std::string onnx_path, const std::string trt_engine_path);
 57 | 
 58 |     int run(const std::string img_path);
 59 | 
 60 | private:
 61 | 
 62 |     TrtLogger logger;
 63 |     std::shared_ptr<nvinfer1::ICudaEngine> engine;
 64 |     bool debug_ = true;
 65 | 
 66 |     int maxworkspace_bit = 28; // 2的28次方表示的trt workspace空间
 67 | 
 68 |     int input_width = 224;
 69 |     int input_height = 224;
 70 |     cv::Scalar mean = (0.485, 0.456, 0.406);
 71 |     cv::Scalar std = (0.229, 0.224, 0.225);
 72 |         
 73 | };
 74 | 
 75 | 
 76 | #define checkRuntime(op) _check_cuda_runtime((op), #op, __FILE__, __LINE__)
 77 | 
 78 | bool _check_cuda_runtime(cudaError_t code, const std::string op, const std::string file, int line){
 79 |     if (code != cudaSuccess){
 80 |         const std::string err_name = cudaGetErrorName(code);
 81 |         const std::string err_message = cudaGetErrorString(code);
 82 |         printf("Runtime Error %s: %d %s failed. \n code = %s, message = %s \n", file, line, op, err_name, err_message);
 83 |         return false;
 84 |     }
 85 |     return true;
 86 | }
 87 | 
 88 | 
 89 | 
 90 | 
 91 | template<typename _T>
 92 | std::shared_ptr<_T> make_nvshared(_T* ptr){
 93 |     return std::shared_ptr<_T>(ptr, [](_T* p){p->destroy();});
 94 | }
 95 | 
 96 | bool file_exists(const std::string path){
 97 |     std::fstream f(path, std::ios::in);
 98 |     return f.good();
 99 | }
100 | 
101 | 
102 | std::vector<float> softmax_cpu(const std::vector<float>& input_data){
103 |     std::vector<float> res(input_data.size(), 0.0);
104 |     std::vector<float> exps(input_data.size(), 0.0);
105 |     float sums =0;
106 |     for(int i = 0; i <input_data.size(); i++){
107 |         exps[i] = exp(input_data[i]);
108 |         sums += exps[i];
109 |     }
110 |     for(int i = 0; i <input_data.size(); i++){
111 |         res[i] = exps[i] / sums;
112 |     }
113 | 
114 |     return res;
115 | }
116 | 
117 | std::unordered_map<int, std::string> load_label_file(const std::string &path){
118 |     std::unordered_map<int, std::string> index2label;
119 |     std::fstream in(path, std::ios::in);
120 |     if (!in.is_open()){
121 |         printf("%s open is failed \n", path);
122 |     }
123 |     std::string line;
124 |     while (getline(in, line)){
125 |         std::vector<std::string> line_item;
126 |         std::stringstream ss(line);
127 |         std::string tmp;
128 |         while (getline(ss, tmp, ' ')){
129 |             line_item.push_back(tmp);
130 |         }
131 |         index2label.insert({std::stoi(line_item[0]), line_item[1]});
132 |     } 
133 |     return index2label;
134 | 
135 | }
136 | 
137 | std::vector<unsigned char> load_engine_data(const std::string& path){
138 |     std::ifstream in(path, std::ios::in|std::ios::binary);
139 |     if (! in.is_open()){
140 |         printf("%s open failed \n", path);
141 |         return {};
142 |     }
143 |     in.seekg(0, std::ios::end); // 对输入文件定位，第一个参数是偏移量，第二个是基地址
144 |     int length = in.tellg(); // 返回当前定位指针的位置，表示输入流的大小。
145 |     std::vector<uint8_t> data;
146 |     if (length > 0){
147 |         in.seekg(0, std::ios::beg);
148 |         data.resize(length);
149 |         in.read((char*)&data[0], length);
150 |     }
151 |     in.close();
152 |     return data;
153 | 
154 | }
155 | 
156 | #endif
157 | 


--------------------------------------------------------------------------------
/trt_inference/labels.txt:
--------------------------------------------------------------------------------
1 | 0 cat
2 | 1 dog
3 | 


--------------------------------------------------------------------------------
/trt_inference/src/TrtCls.cpp:
--------------------------------------------------------------------------------
  1 | #include <iostream>
  2 | #include <fstream>
  3 | #include <numeric>
  4 | #include <string>
  5 | #include <vector>
  6 | 
  7 | #include <TrtCls.hpp>
  8 | #include<NvOnnxParser.h>
  9 | 
 10 | using namespace std;
 11 | 
 12 | 
 13 | TrtCls::TrtCls(){
 14 | 
 15 | }
 16 | 
 17 | 
 18 | TrtCls::~TrtCls(){
 19 | 
 20 | }
 21 | 
 22 | 
 23 | int TrtCls::build_engine(const string onnx_path, const string trt_engine_path){
 24 |     if (file_exists(trt_engine_path)){
 25 |         printf("Engine has existed, there is no need to rebuild \n");
 26 |         return 0;
 27 |     }
 28 | 
 29 |     //builder
 30 |     auto builder = make_nvshared(nvinfer1::createInferBuilder(logger));
 31 |     //config
 32 |     auto config = make_nvshared(builder->createBuilderConfig());
 33 |     //network
 34 |     auto network = make_nvshared(builder->createNetworkV2(1)); // createNetworkV2( 1U << NetworkDefinitionCreationFlag::kEXPLICIT_BATCH)
 35 | 
 36 |     /*
 37 |         enum class NetworkDefinitionCreationFlag : int32_t
 38 |     {
 39 |         //! Dynamic shape support requires that the kEXPLICIT_BATCH flag is set.
 40 |         //! With dynamic shapes, any of the input dimensions can vary at run-time,
 41 |         //! and there are no implicit dimensions in the network specification. This is specified by using the
 42 |         //! wildcard dimension value -1.
 43 |         kEXPLICIT_BATCH = 0, //!< Mark the network to be an explicit batch network
 44 | 
 45 |         //! Setting the network to be an explicit precision network has the following implications:
 46 |         //! 1) Precision of all input tensors to the network have to be specified with ITensor::setType() function
 47 |         //! 2) Precision of all layer output tensors in the network have to be specified using ILayer::setOutputType()
 48 |         //! function
 49 |         //! 3) The builder will not quantize the weights of any layer including those running in lower precision(INT8). It
 50 |         //! will
 51 |         //! simply cast the weights into the required precision.
 52 |         //! 4) Dynamic ranges must not be provided to run the network in int8 mode. Dynamic ranges of each tensor in the
 53 |         //! explicit
 54 |         //! precision network is [-127,127].
 55 |         //! 5) Quantizing and dequantizing activation values between higher (FP32) and lower (INT8) precision
 56 |         //! will be performed using explicit Scale layers with input/output precision set appropriately.
 57 |         kEXPLICIT_PRECISION TRT_DEPRECATED_ENUM = 1, //! <-- Deprecated, used for backward compatibility
 58 |     };
 59 |     */
 60 | 
 61 | 
 62 |     // onnxparser解析结果，并填充到network中
 63 |     auto parser = make_nvshared(nvonnxparser::createParser(*network, logger));
 64 |     if(! parser->parseFromFile((char*)onnx_path.c_str(), 1)){
 65 |         printf("Faile to parse %s \n", onnx_path);
 66 |     } 
 67 |     //设置workspace的最大内存占用为256MB
 68 |     config->setMaxWorkspaceSize(1<<28);
 69 | 
 70 |     auto profile = builder->createOptimizationProfile();
 71 |     auto input_tensor = network->getInput(0);
 72 |     auto input_dims = input_tensor->getDimensions();
 73 | 
 74 |     input_dims.d[0] = 1;
 75 |     profile->setDimensions(input_tensor->getName(), nvinfer1::OptProfileSelector::kMIN, input_dims);
 76 |     profile->setDimensions(input_tensor->getName(), nvinfer1::OptProfileSelector::kOPT, input_dims);
 77 | 
 78 |     int max_BatchSize =8;
 79 |     profile->setDimensions(input_tensor->getName(), nvinfer1::OptProfileSelector::kMAX, input_dims);
 80 |     config->addOptimizationProfile(profile);
 81 | 
 82 |     auto bengine = make_nvshared(builder->buildEngineWithConfig(*network, *config));
 83 | 
 84 |     if(bengine == nullptr){
 85 |         printf("build engine failed \n");
 86 |         return -1;
 87 |     }
 88 | 
 89 |     auto model_data = make_nvshared(bengine->serialize());
 90 |     FILE* f = fopen((char*)trt_engine_path.c_str(), "wb");
 91 |     fwrite(model_data->data(), 1, model_data->size(), f);
 92 |     fclose(f);
 93 | 
 94 |     printf("build engine done. \n");
 95 |     return 0;
 96 | }
 97 | 
 98 | 
 99 | 
100 | 
101 | int TrtCls::init_engine(const string onnx_path, const string trt_engine_path){
102 |     if (0 > build_engine(onnx_path, trt_engine_path)){
103 |         printf("build engine failed \n");
104 |         return -1;
105 |     }
106 |     auto engine_data = load_engine_data(trt_engine_path);
107 |     cout<<"engine size: "<<engine_data.size()<<endl;
108 |     auto runtime = make_nvshared(nvinfer1::createInferRuntime(logger));
109 |     engine = make_nvshared(runtime->deserializeCudaEngine(engine_data.data(), engine_data.size()));
110 |     if(engine == nullptr){
111 |         printf("Deserialize cuda engine failed.\n");
112 |         return -1;
113 |     }
114 |     return 0;
115 | 
116 | 
117 | } 
118 | 
119 | 
120 | int TrtCls::run(const string img_path ){
121 | 
122 |     cudaStream_t stream = nullptr;
123 |     checkRuntime(cudaStreamCreate(&stream));
124 |     auto execution_context = make_nvshared(engine->createExecutionContext());
125 | 
126 |     int input_batch = 1;
127 |     int input_channel = 3;
128 |     int input_height = 224;
129 |     int input_width = 224;
130 |     int input_numel = input_batch * input_channel * input_height * input_width;
131 |     float* input_data_host = nullptr;
132 |     float* input_data_device = nullptr;
133 |     checkRuntime(cudaMallocHost(&input_data_host, input_numel * sizeof(float)));
134 |     checkRuntime(cudaMalloc(&input_data_device, input_numel * sizeof(float)));
135 | 
136 |     ///////////////////////////////////////////////////
137 |     // image to float
138 |     auto image = cv::imread(img_path);
139 |     std::vector<float> mean = {0.406, 0.456, 0.485};
140 |     std::vector<float> _std = {0.225, 0.224, 0.229};
141 |     
142 |     // resize
143 |     cv::resize(image, image, cv::Size(input_width, input_height));
144 |     int image_area = image.cols * image.rows;
145 |     unsigned char* pimage = image.data;
146 |     float* phost_b = input_data_host + image_area * 0;
147 |     float* phost_g = input_data_host + image_area * 1;
148 |     float* phost_r = input_data_host + image_area * 2;
149 |     // BGR2RGB  2tensor
150 |     for(int i = 0; i < image_area; ++i, pimage += 3){
151 |         *phost_r++ = (pimage[0] / 255.0f - mean[0]) / _std[0];
152 |         *phost_g++ = (pimage[1] / 255.0f - mean[1]) / _std[1];
153 |         *phost_b++ = (pimage[2] / 255.0f - mean[2]) / _std[2];
154 |     }
155 |     ///////////////////////////////////////////////////
156 |     checkRuntime(cudaMemcpyAsync(input_data_device, input_data_host, input_numel * sizeof(float), cudaMemcpyHostToDevice, stream));
157 | 
158 |     // 3x3输入，对应3x3输出
159 |     const int num_classes = 2;
160 |     float output_data_host[num_classes];
161 |     float* output_data_device = nullptr;
162 |     checkRuntime(cudaMalloc(&output_data_device, sizeof(output_data_host)));
163 | 
164 |     // 明确当前推理时，使用的数据输入大小
165 |     auto input_dims = execution_context->getBindingDimensions(0);
166 |     input_dims.d[0] = input_batch;
167 | 
168 |     // 设置当前推理时，input大小
169 |     execution_context->setBindingDimensions(0, input_dims);
170 |     float* bindings[] = {input_data_device, output_data_device};
171 |     bool success      = execution_context->enqueueV2((void**)bindings, stream, nullptr);
172 |     checkRuntime(cudaMemcpyAsync(output_data_host, output_data_device, sizeof(output_data_host), cudaMemcpyDeviceToHost, stream));
173 |     checkRuntime(cudaStreamSynchronize(stream));
174 |     vector<float> output_data_host_vector;
175 |     for(int i =0; i< num_classes; i++){
176 |         output_data_host_vector.push_back(output_data_host[i]);
177 |     }
178 |     auto prob = softmax_cpu(output_data_host_vector);
179 |     int predict_label = std::max_element(prob.begin(), prob.end())-prob.begin();  // 确定预测类别的下标
180 |     auto labels = load_label_file("labels.txt");
181 |     auto predict_name = labels[predict_label];
182 |     float confidence  = prob[predict_label];    // 获得预测值的置信度
183 |     printf("Predict: %s, confidence = %f, label = %d\n", predict_name.c_str(), confidence, predict_label);
184 | 
185 |     checkRuntime(cudaStreamDestroy(stream));
186 |     checkRuntime(cudaFreeHost(input_data_host));
187 |     checkRuntime(cudaFree(input_data_device));
188 |     checkRuntime(cudaFree(output_data_device));  
189 | 
190 | }
191 | 
192 | int main(int argc, const char** argv){
193 |     TrtCls trtclsServer;
194 |     const string img_path = argv[1];
195 |     const string onnx_path ="./traced_res50.onnx";
196 |     string trt_engine_path = "./res50_engine.trtmodel";
197 |     if (0 >trtclsServer.init_engine(onnx_path, trt_engine_path)){
198 |         printf("model initialize failed \n");
199 |         return -1;
200 |     }
201 |     if (0 > trtclsServer.run(img_path)){
202 |         printf("%s inference failed \n", img_path);
203 |         return -1;
204 |     }
205 |     return 0;
206 | }
207 | 
208 | 
209 | 


--------------------------------------------------------------------------------
/utils/__init__.py:
--------------------------------------------------------------------------------
1 | from .label_smoothing_pytorch import *
2 | from .warmup_lr import *
3 | from .loss_kd import *
4 | from .util import *
5 | from .lr_scheduler import *
6 | from .opts import *
7 | 


--------------------------------------------------------------------------------
/utils/label_smoothing_pytorch.py:
--------------------------------------------------------------------------------
 1 | 
 2 | import torch.nn as nn
 3 | import torch.nn.functional as F
 4 | 
 5 | class LabelSmoothingCrossEntropy(nn.Module):
 6 |     def __init__(self, eps=0.1, reduction='mean'):
 7 |         super(LabelSmoothingCrossEntropy, self).__init__()
 8 |         self.eps = eps
 9 |         self.reduction = reduction
10 | 
11 |     def forward(self, output, target):
12 |         c = output.size()[-1]
13 |         log_preds = F.log_softmax(output, dim=-1)
14 |         if self.reduction=='sum':
15 |             loss = -log_preds.sum()
16 |         else:
17 |             loss = -log_preds.sum(dim=-1)
18 |             if self.reduction=='mean':
19 |                 loss = loss.mean()
20 |         return loss*self.eps/c + (1-self.eps) * F.nll_loss(log_preds, target, reduction=self.reduction)
21 | 
22 | 
23 | if __name__ == '__main__':
24 |     torch.manual_seed(15)
25 |     criterion = LabelSmoothingCrossEntropy()
26 |     out = torch.randn(20, 10)
27 |     lbs = torch.randint(10, (20,))
28 |     print('out:', out, out.size())
29 |     print('lbs:', lbs, lbs.size())
30 | 
31 |     import torch.nn.functional as F
32 |     
33 |     loss = criterion(out, lbs)
34 |     print('loss:', loss)
35 | 


--------------------------------------------------------------------------------
/utils/loss_kd.py:
--------------------------------------------------------------------------------
 1 | import torch.nn as nn
 2 | import torch.nn.functional as F
 3 | 
 4 | def loss_fn_kd(outputs, labels, teacher_outputs, T, alpha):
 5 |     """
 6 |     Compute the knowledge-distillation (KD) loss given outputs, labels.
 7 |     "Hyperparameters": temperature and alpha
 8 | 
 9 |     NOTE: the KL Divergence for PyTorch comparing the softmaxs of teacher
10 |     and student expects the input tensor to be log probabilities!
11 |     """
12 | 
13 |     KD_loss = nn.KLDivLoss()(F.log_softmax(outputs/T, dim=1),
14 |                              F.softmax(teacher_outputs/T, dim=1)) * (alpha * T * T) + \
15 |               F.cross_entropy(outputs, labels) * (1. - alpha)
16 | 
17 |     return KD_loss


--------------------------------------------------------------------------------
/utils/lr_scheduler.py:
--------------------------------------------------------------------------------
 1 | 
 2 | from torch.optim.lr_scheduler import _LRScheduler, MultiStepLR, CosineAnnealingLR, LambdaLR
 3 | 
 4 | class GradualWarmupScheduler(_LRScheduler):
 5 |     """ Gradually warm-up(increasing) learning rate in optimizer.
 6 |       Proposed in 'Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour'.
 7 |       Args:
 8 |           optimizer (Optimizer): Wrapped optimizer.
 9 |           multiplier: init learning rate = base lr / multiplier
10 |           warmup_epoch: target learning rate is reached at warmup_epoch, gradually
11 |           after_scheduler: after target_epoch, use this scheduler(eg. ReduceLROnPlateau)
12 |       """
13 | 
14 |     def __init__(self, optimizer, multiplier, warmup_epoch, after_scheduler, last_epoch=-1):
15 |         self.multiplier = multiplier
16 |         if self.multiplier <= 1.:
17 |             raise ValueError('multiplier should be greater than 1.')
18 |         self.warmup_epoch = warmup_epoch
19 |         self.after_scheduler = after_scheduler
20 |         self.finished = False
21 |         super().__init__(optimizer, last_epoch=last_epoch)
22 | 
23 |     def get_lr(self):
24 |         if self.last_epoch > self.warmup_epoch:
25 |             return self.after_scheduler.get_lr()
26 |         else:
27 |             return [base_lr / self.multiplier * ((self.multiplier - 1.) * self.last_epoch / self.warmup_epoch + 1.)
28 |                     for base_lr in self.base_lrs]
29 | 
30 |     def step(self, epoch=None):
31 |         if epoch is None:
32 |             epoch = self.last_epoch + 1
33 |         self.last_epoch = epoch
34 |         if epoch > self.warmup_epoch:
35 |             self.after_scheduler.step(epoch - self.warmup_epoch)
36 |         else:
37 |             super(GradualWarmupScheduler, self).step(epoch)
38 | 
39 |     def state_dict(self):
40 |         """Returns the state of the scheduler as a :class:`dict`.
41 | 
42 |         It contains an entry for every variable in self.__dict__ which
43 |         is not the optimizer.
44 |         """
45 | 
46 |         state = {key: value for key, value in self.__dict__.items() if key != 'optimizer' and key != 'after_scheduler'}
47 |         state['after_scheduler'] = self.after_scheduler.state_dict()
48 |         return state
49 | 
50 |     def load_state_dict(self, state_dict):
51 |         """Loads the schedulers state.
52 | 
53 |         Arguments:
54 |             state_dict (dict): scheduler state. Should be an object returned
55 |                 from a call to :meth:`state_dict`.
56 |         """
57 | 
58 |         after_scheduler_state = state_dict.pop('after_scheduler')
59 |         self.__dict__.update(state_dict)
60 |         self.after_scheduler.load_state_dict(after_scheduler_state)
61 | 
62 | 
63 | 
64 |         
65 | def get_scheduler(optimizer, n_iter_per_epoch, args):
66 |     if "cosine" in args.lr_type:
67 |         scheduler = CosineAnnealingLR(
68 |             optimizer=optimizer,
69 |             eta_min=0.00001,
70 |             T_max=(args.epochs - args.warmup_epoch) * n_iter_per_epoch)
71 |     elif "step" in args.lr_type:
72 |         scheduler = MultiStepLR(
73 |             optimizer=optimizer,
74 |             gamma=args.lr_decay_rate,
75 |             milestones=[(m - args.warmup_epoch) * n_iter_per_epoch for m in args.lr_steps])        
76 |         
77 |     else:
78 |         raise NotImplementedError("scheduler not supported:" + args.lr_type)
79 | 
80 | 
81 |     if args.warmup_epoch != 0 :
82 |         scheduler = GradualWarmupScheduler(
83 |             optimizer,
84 |             multiplier=args.warmup_multiplier,
85 |             after_scheduler=scheduler,
86 |             warmup_epoch=args.warmup_epoch * n_iter_per_epoch)
87 | 
88 |     return scheduler


--------------------------------------------------------------------------------
/utils/opts.py:
--------------------------------------------------------------------------------
 1 | # -*- coding:utf-8 -*-
 2 | # @time :2022.07.30
 3 | # @author :lxztju
 4 | # @github : https://github.com/lxztju
 5 | 
 6 | import argparse
 7 | parser = argparse.ArgumentParser(description="PyTorch implementation of image classification")
 8 | 
 9 | # ========================= Data Configs ==========================
10 | parser.add_argument('--num_classes', type=int, default=2, help='the numbers of the image classification task')
11 | parser.add_argument('--input_size', default=224, type=int, help="the input feature dimension for the ")
12 | parser.add_argument('--train_list', default='', type=str, help='the path of training samples text file')
13 | parser.add_argument('--val_list', default='', type=str, help='the path of validation samples text file')
14 | 
15 | 
16 | 
17 | # ========================= Model Configs ==========================
18 | parser.add_argument('--model_name', type=str, default="resnet50")
19 | parser.add_argument('--is_pretrained',default=False, action="store_true", help='using imagenet pretrained model')
20 | parser.add_argument('--tune_from', type=str, default='', help='fine-tune from checkpoint')
21 | parser.add_argument('--focal-loss', default=False, action="store_true")
22 | parser.add_argument('--label-smooth', default=False, action="store_true")
23 | parser.add_argument('--resume', type=str, default='', help='resume from checkpoint')
24 | parser.add_argument('--dropout', type=float, default=0.5)
25 | 
26 | 
27 | # ========================= Learning Configs ==========================
28 | parser.add_argument('--epochs', default=20, type=int, metavar='N',
29 |                     help='number of total epochs to run')
30 | parser.add_argument('-b', '--batch-size', default=64, type=int,
31 |                     metavar='N', help='mini-batch size')
32 | parser.add_argument('--warmup_epoch', type=int, default=5)
33 | parser.add_argument('--lr_decay_rate', type=float, default=0.1)
34 | parser.add_argument('--warmup_multiplier', type=int, default=100)    
35 | 
36 | parser.add_argument('--print-freq', '-p', default=10, type=int,
37 |                     metavar='N', help='print frequency (default: 10)')
38 | 
39 | 
40 | parser.add_argument('--lr', '--learning-rate', default=0.01, type=float,
41 |                     metavar='LR', help='initial learning rate')
42 | parser.add_argument('--lr_type', default='cosine', type=str,
43 |                     metavar='LRtype', help='learning rate type')
44 | parser.add_argument('--lr_steps', default=[50, 100], type=float, nargs="+",
45 |                     metavar='LRSteps', help='epochs to decay learning rate by 10')
46 | 
47 | parser.add_argument('--momentum', default=0.9, type=float, metavar='M',
48 |                     help='momentum')
49 | parser.add_argument('--weight-decay', '--wd', default=5e-4, type=float,
50 |                     metavar='W', help='weight decay (default: 5e-4)')
51 | 
52 | 
53 | # ========================= Monitor Configs ==========================
54 | parser.add_argument('--eval-freq', '-ef', default=5, type=int,
55 |                     metavar='N', help='evaluation frequency (default: 5)')
56 | 
57 | 
58 | # ========================= Runtime Configs ==========================
59 | parser.add_argument('-j', '--workers', default=8, type=int, metavar='N',
60 |                     help='number of data loading workers (default: 8)')
61 | 
62 | 
63 | parser.add_argument('--start-epoch', default=0, type=int, metavar='N',
64 |                     help='manual epoch number (useful on restarts)')
65 | parser.add_argument('--output', type=str, default='./outputs', help='save dir for logs and outputs')
66 | 
67 | 
68 | parser.add_argument('--backend', default='nccl', type=str, help='Pytorch DDP backend')
69 | parser.add_argument('--local_rank', default=-1, type=int, help='DDP local rank')


--------------------------------------------------------------------------------
/utils/util.py:
--------------------------------------------------------------------------------
 1 | import logging
 2 | import torch
 3 | import random 
 4 | from pathlib import Path
 5 | import numpy as np
 6 | 
 7 | 
 8 | def init_logger(log_file=None, rank=-1):
 9 |     '''
10 |     Example:
11 |         >>> init_logger(log_file)
12 |         >>> logger.info("abc'")
13 |     '''
14 |     if isinstance(log_file, Path):
15 |         log_file = str(log_file)
16 |     log_format = logging.Formatter(fmt='%(asctime)s - %(levelname)s - %(name)s -   %(message)s',
17 |                                    datefmt='%Y/%m/%d %H:%M:%S')
18 | 
19 |     logger = logging.getLogger()
20 |     # 优先级 logging.basicConfig < handler.setLevel < logger.setLevel
21 |     if rank in [-1, 0]:
22 |         logger.setLevel(logging.INFO)
23 |     else:
24 |         logger.setLevel(logging.WARNING)
25 |     console_handler = logging.StreamHandler()
26 |     console_handler.setFormatter(log_format)
27 |     logger.handlers = [console_handler]
28 |     if log_file and log_file != '':
29 |         file_handler = logging.FileHandler(log_file, encoding='utf8', mode='a')
30 |         file_handler.setFormatter(log_format)
31 |         logger.addHandler(file_handler)
32 |     return logger
33 | 
34 | 
35 | 
36 | def distributed_concat(tensor, num_total_examples):
37 |     output_tensors = [tensor.clone() for _ in range(torch.distributed.get_world_size())]
38 |     torch.distributed.all_gather(output_tensors, tensor)
39 |     concat = torch.cat(output_tensors, dim=0)
40 |     # truncate the dummy elements added by SequentialDistributedSampler
41 |     return concat[:num_total_examples]
42 | 
43 | 
44 | from contextlib import contextmanager
45 | # 在某个进程中优先执行A操作，其他进程等待其执行完成后再执行A操作
46 | @contextmanager
47 | def torch_distributed_zero_first(local_rank: int):
48 |     """
49 |     Decorator to make all processes in distributed training wait for each local_master to do something.
50 |     """
51 |     if local_rank not in [-1, 0]:
52 |         torch.distributed.barrier()
53 |     yield   #中断后执行上下文代码，然后返回到此处继续往下执行
54 |     if local_rank == 0:
55 |         torch.distributed.barrier()
56 | 
57 | 
58 | def softmax(scores):
59 |     es = np.exp(scores - scores.max(axis=-1)[..., None])
60 |     return es / es.sum(axis=-1)[..., None]
61 | 
62 | 
63 | class AverageMeter(object):
64 |     """Computes and stores the average and current value"""
65 | 
66 |     def __init__(self):
67 |         self.reset()
68 | 
69 |     def reset(self):
70 |         self.val = 0
71 |         self.avg = 0
72 |         self.sum = 0
73 |         self.count = 0
74 | 
75 |     def update(self, val, n=1):
76 |         self.val = val
77 |         self.sum += val * n
78 |         self.count += n
79 |         self.avg = self.sum / self.count


--------------------------------------------------------------------------------
/utils/warmup_lr.py:
--------------------------------------------------------------------------------
 1 | # -*- coding:utf-8 -*-
 2 | # @time :2020.02.09
 3 | # @IDE : pycharm
 4 | # @author :lxztju
 5 | # @github : https://github.com/lxztju
 6 | import numpy as np
 7 | 
 8 | def adjust_learning_rate_step(optimizer, learning_rate_base, gamma, epoch, step_index, iteration, epoch_size):
 9 |     """Sets the learning rate
10 |     # Adapted from PyTorch Imagenet example:
11 |     # https://github.com/pytorch/examples/blob/master/imagenet/main.py
12 |     """
13 |     if epoch < 4:
14 |         lr = 1e-6 + (learning_rate_base-1e-6) * iteration / (epoch_size * 5)
15 |     else:
16 |         lr = learning_rate_base * (gamma ** (step_index))
17 |     for param_group in optimizer.param_groups:
18 |         param_group['lr'] = lr
19 |     return lr
20 | 
21 | 
22 | def adjust_learning_rate_cosine(optimizer, global_step, learning_rate_base, total_steps, warmup_steps):
23 | 
24 |     lr = cosine_decay_with_warmup(global_step,
25 |                              learning_rate_base,
26 |                              total_steps,
27 |                              warmup_learning_rate=0.0,
28 |                              warmup_steps=warmup_steps,
29 |                              hold_base_rate_steps=0)
30 |     for param_group in optimizer.param_groups:
31 |         param_group['lr'] = lr
32 | 
33 |     return lr
34 | 
35 | 
36 | 
37 | def cosine_decay_with_warmup(global_step,
38 |                              learning_rate_base,
39 |                              total_steps,
40 |                              warmup_learning_rate=0.0,
41 |                              warmup_steps=0,
42 |                              hold_base_rate_steps=0):
43 |     """Cosine decay schedule with warm up period.
44 | 
45 |     Cosine annealing learning rate as described in:
46 |       Loshchilov and Hutter, SGDR: Stochastic Gradient Descent with Warm Restarts.
47 |       ICLR 2017. https://arxiv.org/abs/1608.03983
48 |     In this schedule, the learning rate grows linearly from warmup_learning_rate
49 |     to learning_rate_base for warmup_steps, then transitions to a cosine decay
50 |     schedule.
51 | 
52 |     Arguments:
53 |         global_step {int} -- global step.
54 |         learning_rate_base {float} -- base learning rate.
55 |         total_steps {int} -- total number of training steps.
56 | 
57 |     Keyword Arguments:
58 |         warmup_learning_rate {float} -- initial learning rate for warm up. (default: {0.0})
59 |         warmup_steps {int} -- number of warmup steps. (default: {0})
60 |         hold_base_rate_steps {int} -- Optional number of steps to hold base learning rate
61 |                                     before decaying. (default: {0})
62 |     Returns:
63 |       a float representing learning rate.
64 | 
65 |     Raises:
66 |       ValueError: if warmup_learning_rate is larger than learning_rate_base,
67 |         or if warmup_steps is larger than total_steps.
68 |     """
69 | 
70 |     if total_steps < warmup_steps:
71 |         raise ValueError('total_steps must be larger or equal to '
72 |                          'warmup_steps.')
73 |     # if global_step % (2*warmup_steps)==0:
74 |     #     learning_rate = learning_rate_base * 0.1
75 |     learning_rate = 0.3 * learning_rate_base * (1 + np.cos(
76 |         np.pi *
77 |         (global_step - warmup_steps - hold_base_rate_steps
78 |          ) / float(total_steps - warmup_steps - hold_base_rate_steps)))
79 |     if hold_base_rate_steps > 0:
80 |         learning_rate = np.where(global_step > warmup_steps + hold_base_rate_steps,
81 |                                  learning_rate, learning_rate_base)
82 |     if warmup_steps > 0:
83 |         if learning_rate_base < warmup_learning_rate:
84 |             raise ValueError('learning_rate_base must be larger or equal to '
85 |                              'warmup_learning_rate.')
86 |         slope = (learning_rate_base - warmup_learning_rate) / warmup_steps
87 |         warmup_rate = slope * global_step + warmup_learning_rate
88 |         learning_rate = np.where(global_step < warmup_steps, warmup_rate,
89 |                                  learning_rate)
90 |     return np.where(global_step > total_steps, 0.0, learning_rate)
91 | 
92 | 
93 | 


--------------------------------------------------------------------------------
/visualization/Feature_Visualization.py:
--------------------------------------------------------------------------------
  1 | # -*- coding:utf-8 -*-
  2 | # @time :2020/8/15
  3 | # @IDE : pycharm
  4 | # @author :lxztju
  5 | # @github : https://github.com/lxztju
  6 | 
  7 | '''
  8 | 以resnet50为例,进行特征可视化
  9 | 
 10 | 模型的定义来自于torchvision中的定义
 11 | 
 12 | 针对特定的模型需要查找模型的定义,针对所需可视化的网络层的输出,然后导出特定的输出featuremap
 13 | 
 14 | '''
 15 | 
 16 | 
 17 | import torch
 18 | import numpy as np
 19 | import matplotlib.pyplot as plt
 20 | import torch
 21 | import os
 22 | from PIL import Image
 23 | import numpy as np
 24 | 
 25 | import sys 
 26 | sys.path.append("..") 
 27 | import cfg
 28 | from data import get_test_transform
 29 | 
 30 | # 对于给定的一个网络层的输出x,x为numpy格式的array,维度为[0, channels, width, height]
 31 | # %matplotlib inline
 32 | def draw_features(width, height, channels,x,savename):
 33 |     '''
 34 |     x:        输入的array,某一层的网络层输出
 35 |     savename: 特征可视化的保存路径
 36 |     width, height: 分别表示可视化子图的个数,二者乘积等于channels
 37 |     '''
 38 |     fig = plt.figure(figsize=(32,32))
 39 |     fig.subplots_adjust(left=0.05, right=0.95, bottom=0.05, top=0.95, wspace=0.05, hspace=0.05)
 40 |     for i in range(channels):
 41 |         plt.subplot(height,width, i + 1)
 42 |         plt.axis('off')
 43 |         img = x[0, i, :, :]
 44 |         pmin = np.min(img)
 45 |         pmax = np.max(img)
 46 |         img = (img - pmin) / (pmax - pmin + 0.000001)
 47 |         plt.imshow(img, cmap='gray')
 48 | #         print("{}/{}".format(i, channels))
 49 |     fig.savefig(savename, dpi=300)
 50 |     fig.clf()
 51 |     plt.close()
 52 | 
 53 | 
 54 | # 读取模型
 55 | def load_checkpoint(filepath):
 56 |     checkpoint = torch.load(filepath)
 57 |     model = checkpoint['model']  # 提取网络结构
 58 |     model.load_state_dict(checkpoint['model_state_dict'])  # 加载网络权重参数
 59 |     for parameter in model.parameters():
 60 |         parameter.requires_grad = False
 61 |     model.eval()
 62 | #     print(model)
 63 | #     for name in model.state_dict():
 64 | #         print(name)
 65 |     return model
 66 | 
 67 | 
 68 | savepath = './'
 69 | def predict(model):
 70 |     # 读入模型
 71 |     model = load_checkpoint(model)
 72 |     print('..... Finished loading model! ......')
 73 |     ##将模型放置在gpu上运行
 74 |     if torch.cuda.is_available():
 75 |         model.cuda()
 76 | 
 77 |     img = Image.open(img_path).convert('RGB')
 78 | 
 79 |     img = get_test_transform(size=cfg.INPUT_SIZE)(img).unsqueeze(0)
 80 | 
 81 |     if torch.cuda.is_available():
 82 |         img = img.cuda()
 83 |     with torch.no_grad():
 84 |         x = model.conv1(img)
 85 |         draw_features(8, 8, 64,  x.cpu().numpy(), "{}/f1_conv1.png".format(savepath))
 86 | 
 87 | 
 88 | 
 89 | 
 90 | 
 91 | if __name__ == "__main__":
 92 | 
 93 |     trained_model =  '/disk/haihua/weights/resnet50/epoch_39.pth'
 94 |     img_path = './test.png'
 95 | 
 96 |     predict(trained_model)
 97 | 
 98 | 
 99 | 
100 | 
101 | 


--------------------------------------------------------------------------------
/visualization/f1_conv1.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/visualization/f1_conv1.png


--------------------------------------------------------------------------------
/visualization/test.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/lxztju/pytorch_classification/e28cb9f9058abe0b920b85c4740efbb6c135f13f/visualization/test.png


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