├── LICENSE ├── README.md ├── aug_util.py ├── process_wv.py ├── tfr_util.py ├── wv_util.py ├── xView Processing.ipynb └── xview_class_labels.txt /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "[]" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright [yyyy] [name of copyright owner] 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. 202 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | ## xView Data Utilities 2 | 3 | This repository contains data processing scripts for the xView dataset. The script 'process_wv.py' is runnable and processes a folder containing xView imagery along with a groundtruth geojson file to create a TFRecord containing shuffled, chipped, and augmented xView patches in JPEG format. We provide several augmentation functions in 'aug_util.py' for rotating and shifting images and bounding boxes, as well as noise injecting techniques like salt-and-pepper and gaussian blurring. Additionally in 'wv_util.py' we provide several functions for loading, processing, and chipping xView data. 4 | 5 | The Jupyter Notebook provided in this repository interactively illustrates an example xView processing pipeline using the provided utility functions. -------------------------------------------------------------------------------- /aug_util.py: -------------------------------------------------------------------------------- 1 | """ 2 | Copyright 2018 Defense Innovation Unit Experimental 3 | All rights reserved. 4 | 5 | Licensed under the Apache License, Version 2.0 (the "License"); 6 | you may not use this file except in compliance with the License. 7 | You may obtain a copy of the License at 8 | 9 | http://www.apache.org/licenses/LICENSE-2.0 10 | 11 | Unless required by applicable law or agreed to in writing, software 12 | distributed under the License is distributed on an "AS IS" BASIS, 13 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 | See the License for the specific language governing permissions and 15 | limitations under the License. 16 | """ 17 | 18 | import numpy as np 19 | from PIL import Image 20 | import tensorflow as tf 21 | from PIL import Image, ImageDraw 22 | import skimage.filters as filters 23 | 24 | 25 | """ 26 | Image augmentation utilities to be used for processing the dataset. Importantly, these utilities modify 27 | the images as well as their respective bboxes (for example, in rotation). Includes: 28 | rotation, shifting, salt-and-pepper, gaussian blurring. Also includes a 'draw_bboxes' function 29 | for visualizing augmented images and bboxes 30 | """ 31 | 32 | 33 | def rotate_image_and_boxes(img, deg, pivot, boxes): 34 | """ 35 | Rotates an image and corresponding bounding boxes. Bounding box rotations are kept axis-aligned, 36 | so multiples of non 90-degrees changes the area of the bounding box. 37 | 38 | Args: 39 | img: the image to be rotated in array format 40 | deg: an integer representing degree of rotation 41 | pivot: the axis of rotation. By default should be the center of an image, but this can be changed. 42 | boxes: an (N,4) array of boxes for the image 43 | 44 | Output: 45 | Returns the rotated image array along with correspondingly rotated bounding boxes 46 | """ 47 | 48 | if deg < 0: 49 | deg = 360-deg 50 | deg = int(deg) 51 | 52 | angle = 360-deg 53 | padX = [img.shape[0] - pivot[0], pivot[0]] 54 | padY = [img.shape[1] - pivot[1], pivot[1]] 55 | imgP = np.pad(img, [padY, padX, [0,0]], 'constant').astype(np.uint8) 56 | #scipy ndimage rotate takes ~.7 seconds 57 | #imgR = ndimage.rotate(imgP, angle, reshape=False) 58 | #PIL rotate uses ~.01 seconds 59 | imgR = Image.fromarray(imgP).rotate(angle) 60 | imgR = np.array(imgR) 61 | 62 | theta = deg * (np.pi/180) 63 | R = np.array([[np.cos(theta),-np.sin(theta)],[np.sin(theta),np.cos(theta)]]) 64 | # [(cos(theta), -sin(theta))] DOT [xmin, xmax] = [xmin*cos(theta) - ymin*sin(theta), xmax*cos(theta) - ymax*sin(theta)] 65 | # [sin(theta), cos(theta)] [ymin, ymax] [xmin*sin(theta) + ymin*cos(theta), xmax*cos(theta) + ymax*cos(theta)] 66 | 67 | newboxes = [] 68 | for box in boxes: 69 | xmin, ymin, xmax, ymax = box 70 | #The 'x' values are not centered by the x-center (shape[0]/2) 71 | #but rather the y-center (shape[1]/2) 72 | 73 | xmin -= pivot[1] 74 | xmax -= pivot[1] 75 | ymin -= pivot[0] 76 | ymax -= pivot[0] 77 | 78 | bfull = np.array([ [xmin,xmin,xmax,xmax] , [ymin,ymax,ymin,ymax]]) 79 | c = np.dot(R,bfull) 80 | c[0] += pivot[1] 81 | c[0] = np.clip(c[0],0,img.shape[1]) 82 | c[1] += pivot[0] 83 | c[1] = np.clip(c[1],0,img.shape[0]) 84 | 85 | if np.all(c[1] == img.shape[0]) or np.all(c[1] == 0): 86 | c[0] = [0,0,0,0] 87 | if np.all(c[0] == img.shape[1]) or np.all(c[0] == 0): 88 | c[1] = [0,0,0,0] 89 | 90 | newbox = np.array([np.min(c[0]),np.min(c[1]),np.max(c[0]),np.max(c[1])]).astype(np.int64) 91 | 92 | if not (np.all(c[1] == 0) and np.all(c[0] == 0)): 93 | newboxes.append(newbox) 94 | 95 | return imgR[padY[0] : -padY[1], padX[0] : -padX[1]], newboxes 96 | 97 | def shift_image(image,bbox): 98 | """ 99 | Shift an image by a random amount on the x and y axis drawn from discrete 100 | uniform distribution with parameter min(shape/10) 101 | 102 | Args: 103 | image: the image to be shifted in array format 104 | bbox: an (N,4) array of boxes for the image 105 | 106 | Output: 107 | The shifted image and corresponding boxes 108 | """ 109 | shape = image.shape[:2] 110 | maxdelta = min(shape)/10 111 | dx,dy = np.random.randint(-maxdelta,maxdelta,size=(2)) 112 | newimg = np.zeros(image.shape,dtype=np.uint8) 113 | 114 | nb = [] 115 | for box in bbox: 116 | xmin,xmax = np.clip((box[0]+dy,box[2]+dy),0,shape[1]) 117 | ymin,ymax = np.clip((box[1]+dx,box[3]+dx),0,shape[0]) 118 | 119 | #we only add the box if they are not all 0 120 | if not(xmin==0 and xmax ==0 and ymin==0 and ymax ==0): 121 | nb.append([xmin,ymin,xmax,ymax]) 122 | 123 | newimg[max(dx,0):min(image.shape[0],image.shape[0]+dx), 124 | max(dy,0):min(image.shape[1],image.shape[1]+dy)] = \ 125 | image[max(-dx,0):min(image.shape[0],image.shape[0]-dx), 126 | max(-dy,0):min(image.shape[1],image.shape[1]-dy)] 127 | 128 | return newimg, nb 129 | 130 | def salt_and_pepper(img,prob=.005): 131 | """ 132 | Applies salt and pepper noise to an image with given probability for both. 133 | 134 | Args: 135 | img: the image to be augmented in array format 136 | prob: the probability of applying noise to the image 137 | 138 | Output: 139 | Augmented image 140 | """ 141 | 142 | newimg = np.copy(img) 143 | whitemask = np.random.randint(0,int((1-prob)*200),size=img.shape[:2]) 144 | blackmask = np.random.randint(0,int((1-prob)*200),size=img.shape[:2]) 145 | newimg[whitemask==0] = 255 146 | newimg[blackmask==0] = 0 147 | 148 | return newimg 149 | 150 | 151 | def gaussian_blur(img, max_sigma=1.5): 152 | """ 153 | Use a gaussian filter to blur an image 154 | 155 | Args: 156 | img: image to be augmented in array format 157 | max_sigma: the maximum variance for gaussian blurring 158 | 159 | Output: 160 | Augmented image 161 | """ 162 | return filters.gaussian(img,np.random.random()*max_sigma,multichannel=True)*255 163 | 164 | def draw_bboxes(img,boxes): 165 | """ 166 | A helper function to draw bounding box rectangles on images 167 | 168 | Args: 169 | img: image to be drawn on in array format 170 | boxes: An (N,4) array of bounding boxes 171 | 172 | Output: 173 | Image with drawn bounding boxes 174 | """ 175 | source = Image.fromarray(img) 176 | draw = ImageDraw.Draw(source) 177 | w2,h2 = (img.shape[0],img.shape[1]) 178 | 179 | idx = 0 180 | 181 | for b in boxes: 182 | xmin,ymin,xmax,ymax = b 183 | 184 | for j in range(3): 185 | draw.rectangle(((xmin+j, ymin+j), (xmax+j, ymax+j)), outline="red") 186 | return source -------------------------------------------------------------------------------- /process_wv.py: -------------------------------------------------------------------------------- 1 | """ 2 | Copyright 2018 Defense Innovation Unit Experimental 3 | All rights reserved. 4 | 5 | Licensed under the Apache License, Version 2.0 (the "License"); 6 | you may not use this file except in compliance with the License. 7 | You may obtain a copy of the License at 8 | 9 | http://www.apache.org/licenses/LICENSE-2.0 10 | 11 | Unless required by applicable law or agreed to in writing, software 12 | distributed under the License is distributed on an "AS IS" BASIS, 13 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 | See the License for the specific language governing permissions and 15 | limitations under the License. 16 | """ 17 | 18 | 19 | from PIL import Image 20 | import tensorflow as tf 21 | import io 22 | import glob 23 | from tqdm import tqdm 24 | import numpy as np 25 | import logging 26 | import argparse 27 | import os 28 | import json 29 | import wv_util as wv 30 | import tfr_util as tfr 31 | import aug_util as aug 32 | import csv 33 | 34 | """ 35 | A script that processes xView imagery. 36 | Args: 37 | image_folder: A folder path to the directory storing xView .tif files 38 | ie ("xView_data/") 39 | 40 | json_filepath: A file path to the GEOJSON ground truth file 41 | ie ("xView_gt.geojson") 42 | 43 | test_percent (-t): The percentage of input images to use for test set 44 | 45 | suffix (-s): The suffix for output TFRecord files. Default suffix 't1' will output 46 | xview_train_t1.record and xview_test_t1.record 47 | 48 | augment (-a): A boolean value of whether or not to use augmentation 49 | 50 | Outputs: 51 | Writes two files to the current directory containing training and test data in 52 | TFRecord format ('xview_train_SUFFIX.record' and 'xview_test_SUFFIX.record') 53 | """ 54 | 55 | 56 | def get_images_from_filename_array(coords,chips,classes,folder_names,res=(250,250)): 57 | """ 58 | Gathers and chips all images within a given folder at a given resolution. 59 | 60 | Args: 61 | coords: an array of bounding box coordinates 62 | chips: an array of filenames that each coord/class belongs to. 63 | classes: an array of classes for each bounding box 64 | folder_names: a list of folder names containing images 65 | res: an (X,Y) tuple where (X,Y) are (width,height) of each chip respectively 66 | 67 | Output: 68 | images, boxes, classes arrays containing chipped images, bounding boxes, and classes, respectively. 69 | """ 70 | 71 | images =[] 72 | boxes = [] 73 | clses = [] 74 | 75 | k = 0 76 | bi = 0 77 | 78 | for folder in folder_names: 79 | fnames = glob.glob(folder + "*.tif") 80 | fnames.sort() 81 | for fname in tqdm(fnames): 82 | #Needs to be "X.tif" ie ("5.tif") 83 | name = fname.split("\\")[-1] 84 | arr = wv.get_image(fname) 85 | 86 | img,box,cls = wv.chip_image(arr,coords[chips==name],classes[chips==name],res) 87 | 88 | for im in img: 89 | images.append(im) 90 | for b in box: 91 | boxes.append(b) 92 | for c in cls: 93 | clses.append(cls) 94 | k = k + 1 95 | 96 | return images, boxes, clses 97 | 98 | def shuffle_images_and_boxes_classes(im,box,cls): 99 | """ 100 | Shuffles images, boxes, and classes, while keeping relative matching indices 101 | 102 | Args: 103 | im: an array of images 104 | box: an array of bounding box coordinates ([xmin,ymin,xmax,ymax]) 105 | cls: an array of classes 106 | 107 | Output: 108 | Shuffle image, boxes, and classes arrays, respectively 109 | """ 110 | assert len(im) == len(box) 111 | assert len(box) == len(cls) 112 | 113 | perm = np.random.permutation(len(im)) 114 | out_b = {} 115 | out_c = {} 116 | 117 | k = 0 118 | for ind in perm: 119 | out_b[k] = box[ind] 120 | out_c[k] = cls[ind] 121 | k = k + 1 122 | return im[perm], out_b, out_c 123 | 124 | ''' 125 | Datasets 126 | _multires: multiple resolutions. Currently [(500,500),(400,400),(300,300),(200,200)] 127 | _aug: Augmented dataset 128 | ''' 129 | 130 | if __name__ == "__main__": 131 | parser = argparse.ArgumentParser() 132 | parser.add_argument("image_folder", help="Path to folder containing image chips (ie 'Image_Chips/' ") 133 | parser.add_argument("json_filepath", help="Filepath to GEOJSON coordinate file") 134 | parser.add_argument("-t", "--test_percent", type=float, default=0.333, 135 | help="Percent to split into test (ie .25 = test set is 25% total)") 136 | parser.add_argument("-s", "--suffix", type=str, default='t1', 137 | help="Output TFRecord suffix. Default suffix 't1' will output 'xview_train_t1.record' and 'xview_test_t1.record'") 138 | parser.add_argument("-a","--augment", type=bool, default=False, 139 | help="A boolean value whether or not to use augmentation") 140 | args = parser.parse_args() 141 | 142 | logging.basicConfig(level=logging.INFO) 143 | logger = logging.getLogger(__name__) 144 | 145 | #resolutions should be largest -> smallest. We take the number of chips in the largest resolution and make 146 | #sure all future resolutions have less than 1.5times that number of images to prevent chip size imbalance. 147 | #res = [(500,500),(400,400),(300,300),(200,200)] 148 | res = [(300,300)] 149 | 150 | AUGMENT = args.augment 151 | SAVE_IMAGES = False 152 | images = {} 153 | boxes = {} 154 | train_chips = 0 155 | test_chips = 0 156 | 157 | #Parameters 158 | max_chips_per_res = 100000 159 | train_writer = tf.python_io.TFRecordWriter("xview_train_%s.record" % args.suffix) 160 | test_writer = tf.python_io.TFRecordWriter("xview_test_%s.record" % args.suffix) 161 | 162 | coords,chips,classes = wv.get_labels(args.json_filepath) 163 | 164 | for res_ind, it in enumerate(res): 165 | tot_box = 0 166 | logging.info("Res: %s" % str(it)) 167 | ind_chips = 0 168 | 169 | fnames = glob.glob(args.image_folder + "*.tif") 170 | fnames.sort() 171 | 172 | for fname in tqdm(fnames): 173 | #Needs to be "X.tif", ie ("5.tif") 174 | #Be careful!! Depending on OS you may need to change from '/' to '\\'. Use '/' for UNIX and '\\' for windows 175 | name = fname.split("/")[-1] 176 | arr = wv.get_image(fname) 177 | 178 | im,box,classes_final = wv.chip_image(arr,coords[chips==name],classes[chips==name],it) 179 | 180 | #Shuffle images & boxes all at once. Comment out the line below if you don't want to shuffle images 181 | im,box,classes_final = shuffle_images_and_boxes_classes(im,box,classes_final) 182 | split_ind = int(im.shape[0] * args.test_percent) 183 | 184 | for idx, image in enumerate(im): 185 | tf_example = tfr.to_tf_example(image,box[idx],classes_final[idx]) 186 | 187 | #Check to make sure that the TF_Example has valid bounding boxes. 188 | #If there are no valid bounding boxes, then don't save the image to the TFRecord. 189 | float_list_value = tf_example.features.feature['image/object/bbox/xmin'].float_list.value 190 | 191 | if (ind_chips < max_chips_per_res and np.array(float_list_value).any()): 192 | tot_box+=np.array(float_list_value).shape[0] 193 | 194 | if idx < split_ind: 195 | test_writer.write(tf_example.SerializeToString()) 196 | test_chips+=1 197 | else: 198 | train_writer.write(tf_example.SerializeToString()) 199 | train_chips += 1 200 | 201 | ind_chips +=1 202 | 203 | #Make augmentation probability proportional to chip size. Lower chip size = less chance. 204 | #This makes the chip-size imbalance less severe. 205 | prob = np.random.randint(0,np.max(res)) 206 | #for 200x200: p(augment) = 200/500 ; for 300x300: p(augment) = 300/500 ... 207 | 208 | if AUGMENT and prob < it[0]: 209 | 210 | for extra in range(3): 211 | center = np.array([int(image.shape[0]/2),int(image.shape[1]/2)]) 212 | deg = np.random.randint(-10,10) 213 | #deg = np.random.normal()*30 214 | newimg = aug.salt_and_pepper(aug.gaussian_blur(image)) 215 | 216 | #.3 probability for each of shifting vs rotating vs shift(rotate(image)) 217 | p = np.random.randint(0,3) 218 | if p == 0: 219 | newimg,nb = aug.shift_image(newimg,box[idx]) 220 | elif p == 1: 221 | newimg,nb = aug.rotate_image_and_boxes(newimg,deg,center,box[idx]) 222 | elif p == 2: 223 | newimg,nb = aug.rotate_image_and_boxes(newimg,deg,center,box[idx]) 224 | newimg,nb = aug.shift_image(newimg,nb) 225 | 226 | 227 | newimg = (newimg).astype(np.uint8) 228 | 229 | if idx%1000 == 0 and SAVE_IMAGES: 230 | Image.fromarray(newimg).save('process/img_%s_%s_%s.png'%(name,extra,it[0])) 231 | 232 | if len(nb) > 0: 233 | tf_example = tfr.to_tf_example(newimg,nb,classes_final[idx]) 234 | 235 | #Don't count augmented chips for chip indices 236 | if idx < split_ind: 237 | test_writer.write(tf_example.SerializeToString()) 238 | test_chips += 1 239 | else: 240 | train_writer.write(tf_example.SerializeToString()) 241 | train_chips+=1 242 | else: 243 | if SAVE_IMAGES: 244 | aug.draw_bboxes(newimg,nb).save('process/img_nobox_%s_%s_%s.png'%(name,extra,it[0])) 245 | if res_ind == 0: 246 | max_chips_per_res = int(ind_chips * 1.5) 247 | logging.info("Max chips per resolution: %s " % max_chips_per_res) 248 | 249 | logging.info("Tot Box: %d" % tot_box) 250 | logging.info("Chips: %d" % ind_chips) 251 | 252 | logging.info("saved: %d train chips" % train_chips) 253 | logging.info("saved: %d test chips" % test_chips) 254 | train_writer.close() 255 | test_writer.close() -------------------------------------------------------------------------------- /tfr_util.py: -------------------------------------------------------------------------------- 1 | # Original work Copyright 2017 The TensorFlow Authors. All Rights Reserved. 2 | # Modifications Copyright 2018 Defense Innovation Unit Experimental. 3 | # 4 | # Licensed under the Apache License, Version 2.0 (the "License"); 5 | # you may not use this file except in compliance with the License. 6 | # You may obtain a copy of the License at 7 | # 8 | # http://www.apache.org/licenses/LICENSE-2.0 9 | # 10 | # Unless required by applicable law or agreed to in writing, software 11 | # distributed under the License is distributed on an "AS IS" BASIS, 12 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 | # See the License for the specific language governing permissions and 14 | # limitations under the License. 15 | # ============================================================================== 16 | 17 | from PIL import Image 18 | import tensorflow as tf 19 | import io 20 | import numpy as np 21 | 22 | ''' 23 | TensorflowRecord (TFRecord) processing helper functions to be re-used by any scripts 24 | that create or read TFRecord files. 25 | ''' 26 | 27 | def to_tf_example(img, boxes, class_num): 28 | """ 29 | Converts a single image with respective boxes into a TFExample. Multiple TFExamples make up a TFRecord. 30 | 31 | Args: 32 | img: an image array 33 | boxes: an array of bounding boxes for the given image 34 | class_num: an array of class numbers for each bouding box 35 | 36 | Output: 37 | A TFExample containing encoded image data, scaled bounding boxes with classes, and other metadata. 38 | """ 39 | encoded = convertToJpeg(img) 40 | 41 | width = img.shape[0] 42 | height = img.shape[1] 43 | 44 | xmin = [] 45 | ymin = [] 46 | xmax = [] 47 | ymax = [] 48 | classes = [] 49 | classes_text = [] 50 | 51 | for ind,box in enumerate(boxes): 52 | xmin.append(box[0] / width) 53 | ymin.append(box[1] / height) 54 | xmax.append(box[2] / width) 55 | ymax.append(box[3] / height) 56 | classes.append(int(class_num[ind])) 57 | 58 | example = tf.train.Example(features=tf.train.Features(feature={ 59 | 'image/height': int64_feature(height), 60 | 'image/width': int64_feature(width), 61 | 'image/encoded': bytes_feature(encoded), 62 | 'image/format': bytes_feature('jpeg'.encode('utf8')), 63 | 'image/object/bbox/xmin': float_list_feature(xmin), 64 | 'image/object/bbox/xmax': float_list_feature(xmax), 65 | 'image/object/bbox/ymin': float_list_feature(ymin), 66 | 'image/object/bbox/ymax': float_list_feature(ymax), 67 | 'image/object/class/label': int64_list_feature(classes), 68 | })) 69 | 70 | return example 71 | 72 | def convertToJpeg(im): 73 | """ 74 | Converts an image array into an encoded JPEG string. 75 | 76 | Args: 77 | im: an image array 78 | 79 | Output: 80 | an encoded byte string containing the converted JPEG image. 81 | """ 82 | with io.BytesIO() as f: 83 | im = Image.fromarray(im) 84 | im.save(f, format='JPEG') 85 | return f.getvalue() 86 | 87 | def create_tf_record(output_filename, images, boxes): 88 | """ DEPRECIATED 89 | Creates a TFRecord file from examples. 90 | 91 | Args: 92 | output_filename: Path to where output file is saved. 93 | images: an array of images to create a record for 94 | boxes: an array of bounding box coordinates ([xmin,ymin,xmax,ymax]) with the same index as images 95 | """ 96 | writer = tf.python_io.TFRecordWriter(output_filename) 97 | k = 0 98 | for idx, image in enumerate(images): 99 | if idx % 100 == 0: 100 | print('On image %d of %d' %(idx, len(images))) 101 | 102 | tf_example = to_tf_example(image,boxes[idx],fname) 103 | if np.array(tf_example.features.feature['image/object/bbox/xmin'].float_list.value[0]).any(): 104 | writer.write(tf_example.SerializeToString()) 105 | k = k + 1 106 | 107 | print("saved: %d chips" % k) 108 | writer.close() 109 | 110 | ## VARIOUS HELPERS BELOW ## 111 | 112 | def int64_feature(value): 113 | return tf.train.Feature(int64_list=tf.train.Int64List(value=[value])) 114 | 115 | 116 | def int64_list_feature(value): 117 | return tf.train.Feature(int64_list=tf.train.Int64List(value=value)) 118 | 119 | 120 | def bytes_feature(value): 121 | return tf.train.Feature(bytes_list=tf.train.BytesList(value=[value])) 122 | 123 | 124 | def bytes_list_feature(value): 125 | return tf.train.Feature(bytes_list=tf.train.BytesList(value=value)) 126 | 127 | 128 | def float_list_feature(value): 129 | return tf.train.Feature(float_list=tf.train.FloatList(value=value)) -------------------------------------------------------------------------------- /wv_util.py: -------------------------------------------------------------------------------- 1 | """ 2 | Copyright 2018 Defense Innovation Unit Experimental 3 | All rights reserved. 4 | 5 | Licensed under the Apache License, Version 2.0 (the "License"); 6 | you may not use this file except in compliance with the License. 7 | You may obtain a copy of the License at 8 | 9 | http://www.apache.org/licenses/LICENSE-2.0 10 | 11 | Unless required by applicable law or agreed to in writing, software 12 | distributed under the License is distributed on an "AS IS" BASIS, 13 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 | See the License for the specific language governing permissions and 15 | limitations under the License. 16 | """ 17 | 18 | from PIL import Image 19 | import numpy as np 20 | import json 21 | from tqdm import tqdm 22 | 23 | """ 24 | xView processing helper functions for use in data processing. 25 | """ 26 | 27 | def scale(x,range1=(0,0),range2=(0,0)): 28 | """ 29 | Linear scaling for a value x 30 | """ 31 | return range2[0]*(1 - (x-range1[0]) / (range1[1]-range1[0])) + range2[1]*((x-range1[0]) / (range1[1]-range1[0])) 32 | 33 | 34 | def get_image(fname): 35 | """ 36 | Get an image from a filepath in ndarray format 37 | """ 38 | return np.array(Image.open(fname)) 39 | 40 | 41 | def get_labels(fname): 42 | """ 43 | Gets label data from a geojson label file 44 | 45 | Args: 46 | fname: file path to an xView geojson label file 47 | 48 | Output: 49 | Returns three arrays: coords, chips, and classes corresponding to the 50 | coordinates, file-names, and classes for each ground truth. 51 | """ 52 | with open(fname) as f: 53 | data = json.load(f) 54 | 55 | coords = np.zeros((len(data['features']),4)) 56 | chips = np.zeros((len(data['features'])),dtype="object") 57 | classes = np.zeros((len(data['features']))) 58 | 59 | for i in tqdm(range(len(data['features']))): 60 | if data['features'][i]['properties']['bounds_imcoords'] != []: 61 | b_id = data['features'][i]['properties']['image_id'] 62 | val = np.array([int(num) for num in data['features'][i]['properties']['bounds_imcoords'].split(",")]) 63 | chips[i] = b_id 64 | classes[i] = data['features'][i]['properties']['type_id'] 65 | if val.shape[0] != 4: 66 | print("Issues at %d!" % i) 67 | else: 68 | coords[i] = val 69 | else: 70 | chips[i] = 'None' 71 | 72 | return coords, chips, classes 73 | 74 | 75 | def boxes_from_coords(coords): 76 | """ 77 | Processes a coordinate array from a geojson into (xmin,ymin,xmax,ymax) format 78 | 79 | Args: 80 | coords: an array of bounding box coordinates 81 | 82 | Output: 83 | Returns an array of shape (N,4) with coordinates in proper format 84 | """ 85 | nc = np.zeros((coords.shape[0],4)) 86 | for ind in range(coords.shape[0]): 87 | x1,x2 = coords[ind,:,0].min(),coords[ind,:,0].max() 88 | y1,y2 = coords[ind,:,1].min(),coords[ind,:,1].max() 89 | nc[ind] = [x1,y1,x2,y2] 90 | return nc 91 | 92 | 93 | def chip_image(img,coords,classes,shape=(300,300)): 94 | """ 95 | Chip an image and get relative coordinates and classes. Bounding boxes that pass into 96 | multiple chips are clipped: each portion that is in a chip is labeled. For example, 97 | half a building will be labeled if it is cut off in a chip. If there are no boxes, 98 | the boxes array will be [[0,0,0,0]] and classes [0]. 99 | Note: This chip_image method is only tested on xView data-- there are some image manipulations that can mess up different images. 100 | 101 | Args: 102 | img: the image to be chipped in array format 103 | coords: an (N,4) array of bounding box coordinates for that image 104 | classes: an (N,1) array of classes for each bounding box 105 | shape: an (W,H) tuple indicating width and height of chips 106 | 107 | Output: 108 | An image array of shape (M,W,H,C), where M is the number of chips, 109 | W and H are the dimensions of the image, and C is the number of color 110 | channels. Also returns boxes and classes dictionaries for each corresponding chip. 111 | """ 112 | height,width,_ = img.shape 113 | wn,hn = shape 114 | 115 | w_num,h_num = (int(width/wn),int(height/hn)) 116 | images = np.zeros((w_num*h_num,hn,wn,3)) 117 | total_boxes = {} 118 | total_classes = {} 119 | 120 | k = 0 121 | for i in range(w_num): 122 | for j in range(h_num): 123 | x = np.logical_or( np.logical_and((coords[:,0]<((i+1)*wn)),(coords[:,0]>(i*wn))), 124 | np.logical_and((coords[:,2]<((i+1)*wn)),(coords[:,2]>(i*wn)))) 125 | out = coords[x] 126 | y = np.logical_or( np.logical_and((out[:,1]<((j+1)*hn)),(out[:,1]>(j*hn))), 127 | np.logical_and((out[:,3]<((j+1)*hn)),(out[:,3]>(j*hn)))) 128 | outn = out[y] 129 | out = np.transpose(np.vstack((np.clip(outn[:,0]-(wn*i),0,wn), 130 | np.clip(outn[:,1]-(hn*j),0,hn), 131 | np.clip(outn[:,2]-(wn*i),0,wn), 132 | np.clip(outn[:,3]-(hn*j),0,hn)))) 133 | box_classes = classes[x][y] 134 | 135 | if out.shape[0] != 0: 136 | total_boxes[k] = out 137 | total_classes[k] = box_classes 138 | else: 139 | total_boxes[k] = np.array([[0,0,0,0]]) 140 | total_classes[k] = np.array([0]) 141 | 142 | chip = img[hn*j:hn*(j+1),wn*i:wn*(i+1),:3] 143 | images[k]=chip 144 | 145 | k = k + 1 146 | 147 | return images.astype(np.uint8),total_boxes,total_classes 148 | -------------------------------------------------------------------------------- /xview_class_labels.txt: -------------------------------------------------------------------------------- 1 | 11:Fixed-wing Aircraft 2 | 12:Small Aircraft 3 | 13:Cargo Plane 4 | 15:Helicopter 5 | 17:Passenger Vehicle 6 | 18:Small Car 7 | 19:Bus 8 | 20:Pickup Truck 9 | 21:Utility Truck 10 | 23:Truck 11 | 24:Cargo Truck 12 | 25:Truck w/Box 13 | 26:Truck Tractor 14 | 27:Trailer 15 | 28:Truck w/Flatbed 16 | 29:Truck w/Liquid 17 | 32:Crane Truck 18 | 33:Railway Vehicle 19 | 34:Passenger Car 20 | 35:Cargo Car 21 | 36:Flat Car 22 | 37:Tank car 23 | 38:Locomotive 24 | 40:Maritime Vessel 25 | 41:Motorboat 26 | 42:Sailboat 27 | 44:Tugboat 28 | 45:Barge 29 | 47:Fishing Vessel 30 | 49:Ferry 31 | 50:Yacht 32 | 51:Container Ship 33 | 52:Oil Tanker 34 | 53:Engineering Vehicle 35 | 54:Tower crane 36 | 55:Container Crane 37 | 56:Reach Stacker 38 | 57:Straddle Carrier 39 | 59:Mobile Crane 40 | 60:Dump Truck 41 | 61:Haul Truck 42 | 62:Scraper/Tractor 43 | 63:Front loader/Bulldozer 44 | 64:Excavator 45 | 65:Cement Mixer 46 | 66:Ground Grader 47 | 71:Hut/Tent 48 | 72:Shed 49 | 73:Building 50 | 74:Aircraft Hangar 51 | 76:Damaged Building 52 | 77:Facility 53 | 79:Construction Site 54 | 83:Vehicle Lot 55 | 84:Helipad 56 | 86:Storage Tank 57 | 89:Shipping container lot 58 | 91:Shipping Container 59 | 93:Pylon 60 | 94:Tower 61 | --------------------------------------------------------------------------------