File size: 10,373 Bytes
bb14d6a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 | {
"cells": [
{
"cell_type": "markdown",
"id": "ebe0faa7",
"metadata": {},
"source": [
"This notebook prepares the datasets for training of the turtle detection model. First, it goes through the SeaTurtleID2022 dataset and converts the existing masks into the YOLO format needed by Ultralytics. Then it goes through the TurtlesOfSMSRC dataset, loads the masks created in the smsrc_prepare notebook and again, converts the masks to the YOLO format. Finally, the metadata are merged together and are ready to use the segmentation_train script, which first trains on SeaTurtleID2022 (photos below water) and then finetunes on the combined dataset (photos above water were added)."
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a2e66c17",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"import json\n",
"import shutil\n",
"import numpy as np\n",
"import pandas as pd\n",
"from tqdm import tqdm\n",
"from wildlife_datasets.datasets import SeaTurtleID2022, TurtlesOfSMSRC\n",
"from wildlife_datasets.datasets.utils import find_images, parse_bbox_mask\n",
"from wildlife_datasets.splits import ClosedSetSplit\n",
"from turtle_detector import get_index, rle_to_yolo, uncompressed_rle_to_yolo"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "93be7212",
"metadata": {},
"outputs": [],
"source": [
"root_out = f'/data/wildlife_datasets/turtle-detector'\n",
"\n",
"for addition in ['images/train', 'images/val', 'labels/train', 'labels/val']:\n",
" for dataset_name in ['SeaTurtleID2022', 'TurtlesOfSMSRC']:\n",
" os.makedirs(os.path.join(root_out, addition, dataset_name), exist_ok=True)"
]
},
{
"cell_type": "markdown",
"id": "14b3e193",
"metadata": {},
"source": [
"# SeaTurtleID2022"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "4c664fa1",
"metadata": {},
"outputs": [],
"source": [
"dataset_name = 'SeaTurtleID2022'\n",
"root = '/data/wildlife_datasets/data/SeaTurtleID2022'\n",
"\n",
"dataset = SeaTurtleID2022(root)\n",
"if dataset.df['path'].nunique() != len(dataset):\n",
" raise ValueError('path is not unique')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "64391cd5",
"metadata": {},
"outputs": [],
"source": [
"splitter = ClosedSetSplit(0.8)\n",
"idx_train, idx_test = splitter.split(dataset.df)[0]\n",
"idx_train += 1\n",
"idx_test += 1"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "057b036e",
"metadata": {},
"outputs": [],
"source": [
"flipper_categories = {\n",
" '': 0,\n",
" 'front_left': 2,\n",
" 'front_right': 3,\n",
" 'rear_left': 4,\n",
" 'rear_right': 5,\n",
"}\n",
"\n",
"root_ann = f'{root}/turtles-data/data'\n",
"with open(os.path.join(root_ann, 'annotations.json')) as file:\n",
" annotations = json.load(file)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "7ff91b8a",
"metadata": {},
"outputs": [],
"source": [
"for ann_img in tqdm(annotations['images']):\n",
" file_name = os.path.join(root_ann, ann_img['file_name'])\n",
" if ann_img['id'] in idx_train:\n",
" shutil.copy(file_name, f'{root_out}/images/train/{dataset_name}')\n",
" elif ann_img['id'] in idx_test:\n",
" shutil.copy(file_name, f'{root_out}/images/val/{dataset_name}')\n",
" else:\n",
" raise ValueError('Split wrong')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "a5a19846",
"metadata": {},
"outputs": [],
"source": [
"for ann_ann in tqdm(annotations['annotations']):\n",
" if ann_ann['category_id'] == 1:\n",
" category_id = 0\n",
" elif ann_ann['category_id'] == 3:\n",
" category_id = 1\n",
" else:\n",
" location = ann_ann['attributes'].get('location', '')\n",
" category_id = flipper_categories[location]\n",
"\n",
" image_id = ann_ann['image_id']\n",
" rle = ann_ann['segmentation'] \n",
" yolo_segments = uncompressed_rle_to_yolo(rle, class_id=category_id)\n",
" ann_img = annotations['images'][image_id - 1]\n",
" base_name = os.path.basename(ann_img['file_name'])\n",
" base_name = os.path.splitext(base_name)[0] + '.txt'\n",
"\n",
" if image_id != ann_img['id']:\n",
" raise ValueError('Image ids are not ordered')\n",
" if ann_img['id'] in idx_train:\n",
" file_name = f'{root_out}/labels/train/{dataset_name}/{base_name}'\n",
" elif ann_img['id'] in idx_test:\n",
" file_name = f'{root_out}/labels/val/{dataset_name}/{base_name}'\n",
" else:\n",
" raise ValueError('Split wrong')\n",
"\n",
" with open(file_name, 'a') as myfile:\n",
" for yolo_segment in yolo_segments:\n",
" myfile.write(yolo_segment + '\\n')"
]
},
{
"cell_type": "markdown",
"id": "65a8b7ce",
"metadata": {},
"source": [
"# TurtlesOfSMSRC"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "33d88ac1",
"metadata": {},
"outputs": [],
"source": [
"dataset_name = 'TurtlesOfSMSRC'\n",
"root = '/data/wildlife_datasets/TurtlesOfSMSRC'\n",
"\n",
"dataset = TurtlesOfSMSRC(root)\n",
"masks = pd.read_csv(f'{root}/masks.csv')\n",
"masks['mask'] = masks['mask'].apply(parse_bbox_mask)"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e632972b",
"metadata": {},
"outputs": [],
"source": [
"splitter = ClosedSetSplit(0.8)\n",
"idx_train, idx_test = splitter.split(dataset.df)[0]"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "e6acc2bd",
"metadata": {},
"outputs": [],
"source": [
"annotation_categories = {\n",
" 'turtle': 0,\n",
" 'head': 1,\n",
" 'flipper_fl': 2,\n",
" 'flipper_fr': 3,\n",
" 'flipper_rl': 4,\n",
" 'flipper_rr': 5,\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "b3377df4",
"metadata": {},
"outputs": [],
"source": [
"for image_id in tqdm(masks['image_id'].unique()):\n",
" i = get_index(dataset, image_id)\n",
" file_name = os.path.join(root, dataset.metadata.loc[i, 'path'])\n",
" if i in idx_train:\n",
" shutil.copy(file_name, f'{root_out}/images/train/{dataset_name}')\n",
" elif i in idx_test:\n",
" shutil.copy(file_name, f'{root_out}/images/val/{dataset_name}')\n",
" else:\n",
" raise ValueError('Split wrong')"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d1eba0f2",
"metadata": {},
"outputs": [],
"source": [
"for _, mask in tqdm(masks.iterrows(), total=len(masks)):\n",
" category_id = annotation_categories[mask['label_side']]\n",
" image_id = mask['image_id']\n",
" rle = mask['mask'] \n",
" yolo_segments = rle_to_yolo(rle, class_id=category_id)\n",
" i = get_index(dataset, image_id)\n",
"\n",
" base_name = os.path.basename(dataset.metadata.loc[i, 'path'])\n",
" base_name = os.path.splitext(base_name)[0] + '.txt'\n",
"\n",
" if i in idx_train:\n",
" file_name = f'{root_out}/labels/train/{dataset_name}/{base_name}'\n",
" elif i in idx_test:\n",
" file_name = f'{root_out}/labels/val/{dataset_name}/{base_name}'\n",
" else:\n",
" raise ValueError('Split wrong')\n",
"\n",
" with open(file_name, 'a') as myfile:\n",
" for yolo_segment in yolo_segments:\n",
" myfile.write(yolo_segment + '\\n')"
]
},
{
"cell_type": "markdown",
"id": "0b6f6683",
"metadata": {},
"source": [
"# Create metadata"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "d5908b52",
"metadata": {},
"outputs": [],
"source": [
"n_repeat = {\n",
" 'SeaTurtleID2022': 1,\n",
" 'TurtlesOfSMSRC': 30,\n",
"}\n",
"\n",
"# First split and only then oversample to prevent train-test leak\n",
"images = find_images(root_out)\n",
"images = root_out + '/' + images['path'] + '/' + images['file']\n",
"images_train = images[images.str.contains('/train/')]\n",
"images_test = images[images.str.contains('/val/')]\n",
"if len(images_train) + len(images_test) != len(images):\n",
" raise ValueError('The split into train and test images failed.')\n",
"\n",
"# Oversample (even the test set)\n",
"idx_train = []\n",
"idx_test = []\n",
"for dataset_name in ['SeaTurtleID2022', 'TurtlesOfSMSRC']:\n",
" idx_part = list(images_train[images_train.str.contains(dataset_name)].index)\n",
" idx_train += n_repeat[dataset_name] * idx_part\n",
" idx_part = list(images_test[images_test.str.contains(dataset_name)].index)\n",
" idx_test += n_repeat[dataset_name] * idx_part\n",
"images_train = images_train.loc[idx_train]\n",
"images_test = images_test.loc[idx_test]\n",
"\n",
"# Save the oversampled splits\n",
"images_train.to_csv(f'{root_out}/train.txt', header=False, index=False)\n",
"images_test.to_csv(f'{root_out}/val.txt', header=False, index=False)\n",
"for dataset_name in ['SeaTurtleID2022', 'TurtlesOfSMSRC']:\n",
" subset_train = images_train[images_train.str.contains(dataset_name)]\n",
" subset_train.to_csv(f'{root_out}/train_{dataset_name}.txt', header=False, index=False)\n",
" subset_test = images_test[images_test.str.contains(dataset_name)]\n",
" subset_test.to_csv(f'{root_out}/val_{dataset_name}.txt', header=False, index=False)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "sam3",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.12.12"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|