Arulkumar03/Wheat_HEAD_Detection_Counting_ComputerVision_Model
0
1# -*- coding: utf-8 -*-2# Copyright (c) Facebook, Inc. and its affiliates.3 4 5"""6This file registers pre-defined datasets at hard-coded paths, and their metadata.7 8We hard-code metadata for common datasets. This will enable:91. Consistency check when loading the datasets102. Use models on these standard datasets directly and run demos,11 without having to download the dataset annotations12 13We hard-code some paths to the dataset that's assumed to14exist in "./datasets/".15 16Users SHOULD NOT use this file to create new dataset / metadata for new dataset.17To add new dataset, refer to the tutorial "docs/DATASETS.md".18"""19 20import os21 22from detectron2.data import DatasetCatalog, MetadataCatalog23 24from .builtin_meta import ADE20K_SEM_SEG_CATEGORIES, _get_builtin_metadata25from .cityscapes import load_cityscapes_instances, load_cityscapes_semantic26from .cityscapes_panoptic import register_all_cityscapes_panoptic27from .coco import load_sem_seg, register_coco_instances28from .coco_panoptic import register_coco_panoptic, register_coco_panoptic_separated29from .lvis import get_lvis_instances_meta, register_lvis_instances30from .pascal_voc import register_pascal_voc31 32# ==== Predefined datasets and splits for COCO ==========33 34_PREDEFINED_SPLITS_COCO = {}35_PREDEFINED_SPLITS_COCO["coco"] = {36 "coco_2014_train": ("coco/train2014", "coco/annotations/instances_train2014.json"),37 "coco_2014_val": ("coco/val2014", "coco/annotations/instances_val2014.json"),38 "coco_2014_minival": ("coco/val2014", "coco/annotations/instances_minival2014.json"),39 "coco_2014_valminusminival": (40 "coco/val2014",41 "coco/annotations/instances_valminusminival2014.json",42 ),43 "coco_2017_train": ("coco/train2017", "coco/annotations/instances_train2017.json"),44 "coco_2017_val": ("coco/val2017", "coco/annotations/instances_val2017.json"),45 "coco_2017_test": ("coco/test2017", "coco/annotations/image_info_test2017.json"),46 "coco_2017_test-dev": ("coco/test2017", "coco/annotations/image_info_test-dev2017.json"),47 "coco_2017_val_100": ("coco/val2017", "coco/annotations/instances_val2017_100.json"),48}49 50_PREDEFINED_SPLITS_COCO["coco_person"] = {51 "keypoints_coco_2014_train": (52 "coco/train2014",53 "coco/annotations/person_keypoints_train2014.json",54 ),55 "keypoints_coco_2014_val": ("coco/val2014", "coco/annotations/person_keypoints_val2014.json"),56 "keypoints_coco_2014_minival": (57 "coco/val2014",58 "coco/annotations/person_keypoints_minival2014.json",59 ),60 "keypoints_coco_2014_valminusminival": (61 "coco/val2014",62 "coco/annotations/person_keypoints_valminusminival2014.json",63 ),64 "keypoints_coco_2017_train": (65 "coco/train2017",66 "coco/annotations/person_keypoints_train2017.json",67 ),68 "keypoints_coco_2017_val": ("coco/val2017", "coco/annotations/person_keypoints_val2017.json"),69 "keypoints_coco_2017_val_100": (70 "coco/val2017",71 "coco/annotations/person_keypoints_val2017_100.json",72 ),73}74 75 76_PREDEFINED_SPLITS_COCO_PANOPTIC = {77 "coco_2017_train_panoptic": (78 # This is the original panoptic annotation directory79 "coco/panoptic_train2017",80 "coco/annotations/panoptic_train2017.json",81 # This directory contains semantic annotations that are82 # converted from panoptic annotations.83 # It is used by PanopticFPN.84 # You can use the script at detectron2/datasets/prepare_panoptic_fpn.py85 # to create these directories.86 "coco/panoptic_stuff_train2017",87 ),88 "coco_2017_val_panoptic": (89 "coco/panoptic_val2017",90 "coco/annotations/panoptic_val2017.json",91 "coco/panoptic_stuff_val2017",92 ),93 "coco_2017_val_100_panoptic": (94 "coco/panoptic_val2017_100",95 "coco/annotations/panoptic_val2017_100.json",96 "coco/panoptic_stuff_val2017_100",97 ),98}99 100 101def register_all_coco(root):102 for dataset_name, splits_per_dataset in _PREDEFINED_SPLITS_COCO.items():103 for key, (image_root, json_file) in splits_per_dataset.items():104 # Assume pre-defined datasets live in `./datasets`.105 register_coco_instances(106 key,107 _get_builtin_metadata(dataset_name),108 os.path.join(root, json_file) if "://" not in json_file else json_file,109 os.path.join(root, image_root),110 )111 112 for (113 prefix,114 (panoptic_root, panoptic_json, semantic_root),115 ) in _PREDEFINED_SPLITS_COCO_PANOPTIC.items():116 prefix_instances = prefix[: -len("_panoptic")]117 instances_meta = MetadataCatalog.get(prefix_instances)118 image_root, instances_json = instances_meta.image_root, instances_meta.json_file119 # The "separated" version of COCO panoptic segmentation dataset,120 # e.g. used by Panoptic FPN121 register_coco_panoptic_separated(122 prefix,123 _get_builtin_metadata("coco_panoptic_separated"),124 image_root,125 os.path.join(root, panoptic_root),126 os.path.join(root, panoptic_json),127 os.path.join(root, semantic_root),128 instances_json,129 )130 # The "standard" version of COCO panoptic segmentation dataset,131 # e.g. used by Panoptic-DeepLab132 register_coco_panoptic(133 prefix,134 _get_builtin_metadata("coco_panoptic_standard"),135 image_root,136 os.path.join(root, panoptic_root),137 os.path.join(root, panoptic_json),138 instances_json,139 )140 141 142# ==== Predefined datasets and splits for LVIS ==========143 144 145_PREDEFINED_SPLITS_LVIS = {146 "lvis_v1": {147 "lvis_v1_train": ("coco/", "lvis/lvis_v1_train.json"),148 "lvis_v1_val": ("coco/", "lvis/lvis_v1_val.json"),149 "lvis_v1_test_dev": ("coco/", "lvis/lvis_v1_image_info_test_dev.json"),150 "lvis_v1_test_challenge": ("coco/", "lvis/lvis_v1_image_info_test_challenge.json"),151 },152 "lvis_v0.5": {153 "lvis_v0.5_train": ("coco/", "lvis/lvis_v0.5_train.json"),154 "lvis_v0.5_val": ("coco/", "lvis/lvis_v0.5_val.json"),155 "lvis_v0.5_val_rand_100": ("coco/", "lvis/lvis_v0.5_val_rand_100.json"),156 "lvis_v0.5_test": ("coco/", "lvis/lvis_v0.5_image_info_test.json"),157 },158 "lvis_v0.5_cocofied": {159 "lvis_v0.5_train_cocofied": ("coco/", "lvis/lvis_v0.5_train_cocofied.json"),160 "lvis_v0.5_val_cocofied": ("coco/", "lvis/lvis_v0.5_val_cocofied.json"),161 },162}163 164 165def register_all_lvis(root):166 for dataset_name, splits_per_dataset in _PREDEFINED_SPLITS_LVIS.items():167 for key, (image_root, json_file) in splits_per_dataset.items():168 register_lvis_instances(169 key,170 get_lvis_instances_meta(dataset_name),171 os.path.join(root, json_file) if "://" not in json_file else json_file,172 os.path.join(root, image_root),173 )174 175 176# ==== Predefined splits for raw cityscapes images ===========177_RAW_CITYSCAPES_SPLITS = {178 "cityscapes_fine_{task}_train": ("cityscapes/leftImg8bit/train/", "cityscapes/gtFine/train/"),179 "cityscapes_fine_{task}_val": ("cityscapes/leftImg8bit/val/", "cityscapes/gtFine/val/"),180 "cityscapes_fine_{task}_test": ("cityscapes/leftImg8bit/test/", "cityscapes/gtFine/test/"),181}182 183 184def register_all_cityscapes(root):185 for key, (image_dir, gt_dir) in _RAW_CITYSCAPES_SPLITS.items():186 meta = _get_builtin_metadata("cityscapes")187 image_dir = os.path.join(root, image_dir)188 gt_dir = os.path.join(root, gt_dir)189 190 inst_key = key.format(task="instance_seg")191 DatasetCatalog.register(192 inst_key,193 lambda x=image_dir, y=gt_dir: load_cityscapes_instances(194 x, y, from_json=True, to_polygons=True195 ),196 )197 MetadataCatalog.get(inst_key).set(198 image_dir=image_dir, gt_dir=gt_dir, evaluator_type="cityscapes_instance", **meta199 )200 201 sem_key = key.format(task="sem_seg")202 DatasetCatalog.register(203 sem_key, lambda x=image_dir, y=gt_dir: load_cityscapes_semantic(x, y)204 )205 MetadataCatalog.get(sem_key).set(206 image_dir=image_dir,207 gt_dir=gt_dir,208 evaluator_type="cityscapes_sem_seg",209 ignore_label=255,210 **meta,211 )212 213 214# ==== Predefined splits for PASCAL VOC ===========215def register_all_pascal_voc(root):216 SPLITS = [217 ("voc_2007_trainval", "VOC2007", "trainval"),218 ("voc_2007_train", "VOC2007", "train"),219 ("voc_2007_val", "VOC2007", "val"),220 ("voc_2007_test", "VOC2007", "test"),221 ("voc_2012_trainval", "VOC2012", "trainval"),222 ("voc_2012_train", "VOC2012", "train"),223 ("voc_2012_val", "VOC2012", "val"),224 ]225 for name, dirname, split in SPLITS:226 year = 2007 if "2007" in name else 2012227 register_pascal_voc(name, os.path.join(root, dirname), split, year)228 MetadataCatalog.get(name).evaluator_type = "pascal_voc"229 230 231def register_all_ade20k(root):232 root = os.path.join(root, "ADEChallengeData2016")233 for name, dirname in [("train", "training"), ("val", "validation")]:234 image_dir = os.path.join(root, "images", dirname)235 gt_dir = os.path.join(root, "annotations_detectron2", dirname)236 name = f"ade20k_sem_seg_{name}"237 DatasetCatalog.register(238 name, lambda x=image_dir, y=gt_dir: load_sem_seg(y, x, gt_ext="png", image_ext="jpg")239 )240 MetadataCatalog.get(name).set(241 stuff_classes=ADE20K_SEM_SEG_CATEGORIES[:],242 image_root=image_dir,243 sem_seg_root=gt_dir,244 evaluator_type="sem_seg",245 ignore_label=255,246 )247 248 249# True for open source;250# Internally at fb, we register them elsewhere251if __name__.endswith(".builtin"):252 # Assume pre-defined datasets live in `./datasets`.253 _root = os.path.expanduser(os.getenv("DETECTRON2_DATASETS", "datasets"))254 register_all_coco(_root)255 register_all_lvis(_root)256 register_all_cityscapes(_root)257 register_all_cityscapes_panoptic(_root)258 register_all_pascal_voc(_root)259 register_all_ade20k(_root)260