From 627e00cb14de3ef043327b3c7881597ccf83e691 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Haian=20Huang=28=E6=B7=B1=E5=BA=A6=E7=9C=B8=29?= <1286304229@qq.com> Date: Mon, 6 Nov 2023 16:29:19 +0800 Subject: [PATCH] Support ODinW and evaluate (#11105) --- ...ss_swin-t_a_fpn_dyhead_pretrain_odinw13.py | 338 ++++++++ ...ss_swin-t_a_fpn_dyhead_pretrain_odinw35.py | 771 ++++++++++++++++++ configs/odinw/override_category.py | 96 +++ docs/zh_cn/user_guides/dataset_prepare.md | 55 ++ mmdet/datasets/base_det_dataset.py | 7 + mmdet/datasets/coco.py | 1 + mmdet/datasets/dataset_wrappers.py | 8 + mmdet/evaluation/__init__.py | 1 + mmdet/evaluation/evaluator/__init__.py | 4 + .../evaluator/multi_datasets_evaluator.py | 111 +++ mmdet/models/detectors/glip.py | 81 +- 11 files changed, 1452 insertions(+), 21 deletions(-) create mode 100644 configs/odinw/glip_atss_swin-t_a_fpn_dyhead_pretrain_odinw13.py create mode 100644 configs/odinw/glip_atss_swin-t_a_fpn_dyhead_pretrain_odinw35.py create mode 100644 configs/odinw/override_category.py create mode 100644 mmdet/evaluation/evaluator/__init__.py create mode 100644 mmdet/evaluation/evaluator/multi_datasets_evaluator.py diff --git a/configs/odinw/glip_atss_swin-t_a_fpn_dyhead_pretrain_odinw13.py b/configs/odinw/glip_atss_swin-t_a_fpn_dyhead_pretrain_odinw13.py new file mode 100644 index 00000000000..6c2cc0c6f09 --- /dev/null +++ b/configs/odinw/glip_atss_swin-t_a_fpn_dyhead_pretrain_odinw13.py @@ -0,0 +1,338 @@ +_base_ = '../glip/glip_atss_swin-t_a_fpn_dyhead_pretrain_obj365.py' + +dataset_type = 'CocoDataset' +data_root = 'data/odinw/' + +base_test_pipeline = _base_.test_pipeline +base_test_pipeline[-1]['meta_keys'] = ('img_id', 'img_path', 'ori_shape', + 'img_shape', 'scale_factor', 'text', + 'custom_entities', 'caption_prompt') + +# ---------------------1 AerialMaritimeDrone---------------------# +class_name = ('boat', 'car', 'dock', 'jetski', 'lift') +metainfo = dict(classes=class_name) +_data_root = data_root + 'AerialMaritimeDrone/large/' +dataset_AerialMaritimeDrone = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + test_mode=True, + pipeline=base_test_pipeline, + return_classes=True) +val_evaluator_AerialMaritimeDrone = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------2 Aquarium---------------------# +class_name = ('fish', 'jellyfish', 'penguin', 'puffin', 'shark', 'starfish', + 'stingray') +metainfo = dict(classes=class_name) +_data_root = data_root + 'Aquarium/Aquarium Combined.v2-raw-1024.coco/' + +caption_prompt = None +# caption_prompt = { +# 'penguin': { +# 'suffix': ', which is black and white' +# }, +# 'puffin': { +# 'suffix': ' with orange beaks' +# }, +# 'stingray': { +# 'suffix': ' which is flat and round' +# }, +# } +dataset_Aquarium = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + caption_prompt=caption_prompt, + test_mode=True, + return_classes=True) +val_evaluator_Aquarium = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------3 CottontailRabbits---------------------# +class_name = ('Cottontail-Rabbit', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'CottontailRabbits/' + +caption_prompt = None +# caption_prompt = {'Cottontail-Rabbit': {'name': 'rabbit'}} + +dataset_CottontailRabbits = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + caption_prompt=caption_prompt, + test_mode=True, + return_classes=True) +val_evaluator_CottontailRabbits = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------4 EgoHands---------------------# +class_name = ('hand', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'EgoHands/generic/' + +caption_prompt = None +# caption_prompt = {'hand': {'suffix': ' of a person'}} + +dataset_EgoHands = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + caption_prompt=caption_prompt, + test_mode=True, + return_classes=True) +val_evaluator_EgoHands = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------5 NorthAmericaMushrooms---------------------# +class_name = ('CoW', 'chanterelle') +metainfo = dict(classes=class_name) +_data_root = data_root + 'NorthAmericaMushrooms/North American Mushrooms.v1-416x416.coco/' # noqa + +caption_prompt = None +# caption_prompt = { +# 'CoW': { +# 'name': 'flat mushroom' +# }, +# 'chanterelle': { +# 'name': 'yellow mushroom' +# } +# } + +dataset_NorthAmericaMushrooms = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + caption_prompt=caption_prompt, + test_mode=True, + return_classes=True) +val_evaluator_NorthAmericaMushrooms = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------6 Packages---------------------# +class_name = ('package', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'Packages/Raw/' + +caption_prompt = None +# caption_prompt = { +# 'package': { +# 'prefix': 'there is a ', +# 'suffix': ' on the porch' +# } +# } + +dataset_Packages = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + caption_prompt=caption_prompt, + test_mode=True, + return_classes=True) +val_evaluator_Packages = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------7 PascalVOC---------------------# +class_name = ('aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus', 'car', + 'cat', 'chair', 'cow', 'diningtable', 'dog', 'horse', + 'motorbike', 'person', 'pottedplant', 'sheep', 'sofa', 'train', + 'tvmonitor') +metainfo = dict(classes=class_name) +_data_root = data_root + 'PascalVOC/' +dataset_PascalVOC = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_PascalVOC = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------8 pistols---------------------# +class_name = ('pistol', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'pistols/export/' +dataset_pistols = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='val_annotations_without_background.json', + data_prefix=dict(img=''), + pipeline=base_test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_pistols = dict( + type='CocoMetric', + ann_file=_data_root + 'val_annotations_without_background.json', + metric='bbox') + +# ---------------------9 pothole---------------------# +class_name = ('pothole', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'pothole/' + +caption_prompt = None +# caption_prompt = { +# 'pothole': { +# 'prefix': 'there are some ', +# 'name': 'holes', +# 'suffix': ' on the road' +# } +# } + +dataset_pothole = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_pothole = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------10 Raccoon---------------------# +class_name = ('raccoon', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'Raccoon/Raccoon.v2-raw.coco/' +dataset_Raccoon = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_Raccoon = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------11 ShellfishOpenImages---------------------# +class_name = ('Crab', 'Lobster', 'Shrimp') +metainfo = dict(classes=class_name) +_data_root = data_root + 'ShellfishOpenImages/raw/' +dataset_ShellfishOpenImages = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_ShellfishOpenImages = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------12 thermalDogsAndPeople---------------------# +class_name = ('dog', 'person') +metainfo = dict(classes=class_name) +_data_root = data_root + 'thermalDogsAndPeople/' +dataset_thermalDogsAndPeople = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_thermalDogsAndPeople = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------13 VehiclesOpenImages---------------------# +class_name = ('Ambulance', 'Bus', 'Car', 'Motorcycle', 'Truck') +metainfo = dict(classes=class_name) +_data_root = data_root + 'VehiclesOpenImages/416x416/' +dataset_VehiclesOpenImages = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=base_test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_VehiclesOpenImages = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# --------------------- Config---------------------# +dataset_prefixes = [ + 'AerialMaritimeDrone', 'Aquarium', 'CottontailRabbits', 'EgoHands', + 'NorthAmericaMushrooms', 'Packages', 'PascalVOC', 'pistols', 'pothole', + 'Raccoon', 'ShellfishOpenImages', 'thermalDogsAndPeople', + 'VehiclesOpenImages' +] +datasets = [ + dataset_AerialMaritimeDrone, dataset_Aquarium, dataset_CottontailRabbits, + dataset_EgoHands, dataset_NorthAmericaMushrooms, dataset_Packages, + dataset_PascalVOC, dataset_pistols, dataset_pothole, dataset_Raccoon, + dataset_ShellfishOpenImages, dataset_thermalDogsAndPeople, + dataset_VehiclesOpenImages +] +metrics = [ + val_evaluator_AerialMaritimeDrone, val_evaluator_Aquarium, + val_evaluator_CottontailRabbits, val_evaluator_EgoHands, + val_evaluator_NorthAmericaMushrooms, val_evaluator_Packages, + val_evaluator_PascalVOC, val_evaluator_pistols, val_evaluator_pothole, + val_evaluator_Raccoon, val_evaluator_ShellfishOpenImages, + val_evaluator_thermalDogsAndPeople, val_evaluator_VehiclesOpenImages +] + +# -------------------------------------------------# +val_dataloader = dict( + dataset=dict(_delete_=True, type='ConcatDataset', datasets=datasets)) +test_dataloader = val_dataloader + +val_evaluator = dict( + _delete_=True, + type='MultiDatasetsEvaluator', + metrics=metrics, + dataset_prefixes=dataset_prefixes) +test_evaluator = val_evaluator diff --git a/configs/odinw/glip_atss_swin-t_a_fpn_dyhead_pretrain_odinw35.py b/configs/odinw/glip_atss_swin-t_a_fpn_dyhead_pretrain_odinw35.py new file mode 100644 index 00000000000..292c5ec1639 --- /dev/null +++ b/configs/odinw/glip_atss_swin-t_a_fpn_dyhead_pretrain_odinw35.py @@ -0,0 +1,771 @@ +_base_ = '../glip/glip_atss_swin-t_a_fpn_dyhead_pretrain_obj365.py' + +dataset_type = 'CocoDataset' +data_root = 'data/odinw/' + +# ---------------------1 AerialMaritimeDrone_large---------------------# +class_name = ('boat', 'car', 'dock', 'jetski', 'lift') +metainfo = dict(classes=class_name) +_data_root = data_root + 'AerialMaritimeDrone/large/' +dataset_AerialMaritimeDrone_large = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_AerialMaritimeDrone_large = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------2 AerialMaritimeDrone_tiled---------------------# +class_name = ('boat', 'car', 'dock', 'jetski', 'lift') +metainfo = dict(classes=class_name) +_data_root = data_root + 'AerialMaritimeDrone/tiled/' +dataset_AerialMaritimeDrone_tiled = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_AerialMaritimeDrone_tiled = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------3 AmericanSignLanguageLetters---------------------# +class_name = ('A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', + 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z') +metainfo = dict(classes=class_name) +_data_root = data_root + 'AmericanSignLanguageLetters/American Sign Language Letters.v1-v1.coco/' # noqa +dataset_AmericanSignLanguageLetters = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_AmericanSignLanguageLetters = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------4 Aquarium---------------------# +class_name = ('fish', 'jellyfish', 'penguin', 'puffin', 'shark', 'starfish', + 'stingray') +metainfo = dict(classes=class_name) +_data_root = data_root + 'Aquarium/Aquarium Combined.v2-raw-1024.coco/' +dataset_Aquarium = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_Aquarium = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------5 BCCD---------------------# +class_name = ('Platelets', 'RBC', 'WBC') +metainfo = dict(classes=class_name) +_data_root = data_root + 'BCCD/BCCD.v3-raw.coco/' +dataset_BCCD = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_BCCD = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------6 boggleBoards---------------------# +class_name = ('Q', 'a', 'an', 'b', 'c', 'd', 'e', 'er', 'f', 'g', 'h', 'he', + 'i', 'in', 'j', 'k', 'l', 'm', 'n', 'o', 'o ', 'p', 'q', 'qu', + 'r', 's', 't', 't\\', 'th', 'u', 'v', 'w', 'wild', 'x', 'y', 'z') +metainfo = dict(classes=class_name) +_data_root = data_root + 'boggleBoards/416x416AutoOrient/export/' +dataset_boggleBoards = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='val_annotations_without_background.json', + data_prefix=dict(img=''), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_boggleBoards = dict( + type='CocoMetric', + ann_file=_data_root + 'val_annotations_without_background.json', + metric='bbox') + +# ---------------------7 brackishUnderwater---------------------# +class_name = ('crab', 'fish', 'jellyfish', 'shrimp', 'small_fish', 'starfish') +metainfo = dict(classes=class_name) +_data_root = data_root + 'brackishUnderwater/960x540/' +dataset_brackishUnderwater = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_brackishUnderwater = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------8 ChessPieces---------------------# +class_name = (' ', 'black bishop', 'black king', 'black knight', 'black pawn', + 'black queen', 'black rook', 'white bishop', 'white king', + 'white knight', 'white pawn', 'white queen', 'white rook') +metainfo = dict(classes=class_name) +_data_root = data_root + 'ChessPieces/Chess Pieces.v23-raw.coco/' +dataset_ChessPieces = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/new_annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_ChessPieces = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/new_annotations_without_background.json', + metric='bbox') + +# ---------------------9 CottontailRabbits---------------------# +class_name = ('Cottontail-Rabbit', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'CottontailRabbits/' +dataset_CottontailRabbits = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_CottontailRabbits = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------10 dice---------------------# +class_name = ('1', '2', '3', '4', '5', '6') +metainfo = dict(classes=class_name) +_data_root = data_root + 'dice/mediumColor/export/' +dataset_dice = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='val_annotations_without_background.json', + data_prefix=dict(img=''), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_dice = dict( + type='CocoMetric', + ann_file=_data_root + 'val_annotations_without_background.json', + metric='bbox') + +# ---------------------11 DroneControl---------------------# +class_name = ('follow', 'follow_hand', 'land', 'land_hand', 'null', 'object', + 'takeoff', 'takeoff-hand') +metainfo = dict(classes=class_name) +_data_root = data_root + 'DroneControl/Drone Control.v3-raw.coco/' +dataset_DroneControl = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_DroneControl = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------12 EgoHands_generic---------------------# +class_name = ('hand', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'EgoHands/generic/' +dataset_EgoHands_generic = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_EgoHands_generic = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------13 EgoHands_specific---------------------# +class_name = ('myleft', 'myright', 'yourleft', 'yourright') +metainfo = dict(classes=class_name) +_data_root = data_root + 'EgoHands/specific/' +dataset_EgoHands_specific = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_EgoHands_specific = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------14 HardHatWorkers---------------------# +class_name = ('head', 'helmet', 'person') +metainfo = dict(classes=class_name) +_data_root = data_root + 'HardHatWorkers/raw/' +dataset_HardHatWorkers = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_HardHatWorkers = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------15 MaskWearing---------------------# +class_name = ('mask', 'no-mask') +metainfo = dict(classes=class_name) +_data_root = data_root + 'MaskWearing/raw/' +dataset_MaskWearing = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_MaskWearing = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------16 MountainDewCommercial---------------------# +class_name = ('bottle', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'MountainDewCommercial/' +dataset_MountainDewCommercial = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_MountainDewCommercial = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------17 NorthAmericaMushrooms---------------------# +class_name = ('flat mushroom', 'yellow mushroom') +metainfo = dict(classes=class_name) +_data_root = data_root + 'NorthAmericaMushrooms/North American Mushrooms.v1-416x416.coco/' # noqa +dataset_NorthAmericaMushrooms = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/new_annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_NorthAmericaMushrooms = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/new_annotations_without_background.json', + metric='bbox') + +# ---------------------18 openPoetryVision---------------------# +class_name = ('American Typewriter', 'Andale Mono', 'Apple Chancery', 'Arial', + 'Avenir', 'Baskerville', 'Big Caslon', 'Bradley Hand', + 'Brush Script MT', 'Chalkboard', 'Comic Sans MS', 'Copperplate', + 'Courier', 'Didot', 'Futura', 'Geneva', 'Georgia', 'Gill Sans', + 'Helvetica', 'Herculanum', 'Impact', 'Kefa', 'Lucida Grande', + 'Luminari', 'Marker Felt', 'Menlo', 'Monaco', 'Noteworthy', + 'Optima', 'PT Sans', 'PT Serif', 'Palatino', 'Papyrus', + 'Phosphate', 'Rockwell', 'SF Pro', 'SignPainter', 'Skia', + 'Snell Roundhand', 'Tahoma', 'Times New Roman', 'Trebuchet MS', + 'Verdana') +metainfo = dict(classes=class_name) +_data_root = data_root + 'openPoetryVision/512x512/' +dataset_openPoetryVision = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_openPoetryVision = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------19 OxfordPets_by_breed---------------------# +class_name = ('cat-Abyssinian', 'cat-Bengal', 'cat-Birman', 'cat-Bombay', + 'cat-British_Shorthair', 'cat-Egyptian_Mau', 'cat-Maine_Coon', + 'cat-Persian', 'cat-Ragdoll', 'cat-Russian_Blue', 'cat-Siamese', + 'cat-Sphynx', 'dog-american_bulldog', + 'dog-american_pit_bull_terrier', 'dog-basset_hound', + 'dog-beagle', 'dog-boxer', 'dog-chihuahua', + 'dog-english_cocker_spaniel', 'dog-english_setter', + 'dog-german_shorthaired', 'dog-great_pyrenees', 'dog-havanese', + 'dog-japanese_chin', 'dog-keeshond', 'dog-leonberger', + 'dog-miniature_pinscher', 'dog-newfoundland', 'dog-pomeranian', + 'dog-pug', 'dog-saint_bernard', 'dog-samoyed', + 'dog-scottish_terrier', 'dog-shiba_inu', + 'dog-staffordshire_bull_terrier', 'dog-wheaten_terrier', + 'dog-yorkshire_terrier') +metainfo = dict(classes=class_name) +_data_root = data_root + 'OxfordPets/by-breed/' # noqa +dataset_OxfordPets_by_breed = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_OxfordPets_by_breed = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------20 OxfordPets_by_species---------------------# +class_name = ('cat', 'dog') +metainfo = dict(classes=class_name) +_data_root = data_root + 'OxfordPets/by-species/' # noqa +dataset_OxfordPets_by_species = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_OxfordPets_by_species = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------21 PKLot---------------------# +class_name = ('space-empty', 'space-occupied') +metainfo = dict(classes=class_name) +_data_root = data_root + 'PKLot/640/' # noqa +dataset_PKLot = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_PKLot = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------22 Packages---------------------# +class_name = ('package', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'Packages/Raw/' +dataset_Packages = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_Packages = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------23 PascalVOC---------------------# +class_name = ('aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus', 'car', + 'cat', 'chair', 'cow', 'diningtable', 'dog', 'horse', + 'motorbike', 'person', 'pottedplant', 'sheep', 'sofa', 'train', + 'tvmonitor') +metainfo = dict(classes=class_name) +_data_root = data_root + 'PascalVOC/' +dataset_PascalVOC = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_PascalVOC = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------24 pistols---------------------# +class_name = ('pistol', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'pistols/export/' +dataset_pistols = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='val_annotations_without_background.json', + data_prefix=dict(img=''), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_pistols = dict( + type='CocoMetric', + ann_file=_data_root + 'val_annotations_without_background.json', + metric='bbox') + +# ---------------------25 plantdoc---------------------# +class_name = ('Apple Scab Leaf', 'Apple leaf', 'Apple rust leaf', + 'Bell_pepper leaf', 'Bell_pepper leaf spot', 'Blueberry leaf', + 'Cherry leaf', 'Corn Gray leaf spot', 'Corn leaf blight', + 'Corn rust leaf', 'Peach leaf', 'Potato leaf', + 'Potato leaf early blight', 'Potato leaf late blight', + 'Raspberry leaf', 'Soyabean leaf', 'Soybean leaf', + 'Squash Powdery mildew leaf', 'Strawberry leaf', + 'Tomato Early blight leaf', 'Tomato Septoria leaf spot', + 'Tomato leaf', 'Tomato leaf bacterial spot', + 'Tomato leaf late blight', 'Tomato leaf mosaic virus', + 'Tomato leaf yellow virus', 'Tomato mold leaf', + 'Tomato two spotted spider mites leaf', 'grape leaf', + 'grape leaf black rot') +metainfo = dict(classes=class_name) +_data_root = data_root + 'plantdoc/416x416/' +dataset_plantdoc = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_plantdoc = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------26 pothole---------------------# +class_name = ('pothole', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'pothole/' +dataset_pothole = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_pothole = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------27 Raccoon---------------------# +class_name = ('raccoon', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'Raccoon/Raccoon.v2-raw.coco/' +dataset_Raccoon = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_Raccoon = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------28 selfdrivingCar---------------------# +class_name = ('biker', 'car', 'pedestrian', 'trafficLight', + 'trafficLight-Green', 'trafficLight-GreenLeft', + 'trafficLight-Red', 'trafficLight-RedLeft', + 'trafficLight-Yellow', 'trafficLight-YellowLeft', 'truck') +metainfo = dict(classes=class_name) +_data_root = data_root + 'selfdrivingCar/fixedLarge/export/' +dataset_selfdrivingCar = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='val_annotations_without_background.json', + data_prefix=dict(img=''), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_selfdrivingCar = dict( + type='CocoMetric', + ann_file=_data_root + 'val_annotations_without_background.json', + metric='bbox') + +# ---------------------29 ShellfishOpenImages---------------------# +class_name = ('Crab', 'Lobster', 'Shrimp') +metainfo = dict(classes=class_name) +_data_root = data_root + 'ShellfishOpenImages/raw/' +dataset_ShellfishOpenImages = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_ShellfishOpenImages = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------30 ThermalCheetah---------------------# +class_name = ('cheetah', 'human') +metainfo = dict(classes=class_name) +_data_root = data_root + 'ThermalCheetah/' +dataset_ThermalCheetah = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_ThermalCheetah = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------31 thermalDogsAndPeople---------------------# +class_name = ('dog', 'person') +metainfo = dict(classes=class_name) +_data_root = data_root + 'thermalDogsAndPeople/' +dataset_thermalDogsAndPeople = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_thermalDogsAndPeople = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------32 UnoCards---------------------# +class_name = ('0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '10', '11', + '12', '13', '14') +metainfo = dict(classes=class_name) +_data_root = data_root + 'UnoCards/raw/' +dataset_UnoCards = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_UnoCards = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------33 VehiclesOpenImages---------------------# +class_name = ('Ambulance', 'Bus', 'Car', 'Motorcycle', 'Truck') +metainfo = dict(classes=class_name) +_data_root = data_root + 'VehiclesOpenImages/416x416/' +dataset_VehiclesOpenImages = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_VehiclesOpenImages = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------34 WildfireSmoke---------------------# +class_name = ('smoke', ) +metainfo = dict(classes=class_name) +_data_root = data_root + 'WildfireSmoke/' +dataset_WildfireSmoke = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_WildfireSmoke = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# ---------------------35 websiteScreenshots---------------------# +class_name = ('button', 'field', 'heading', 'iframe', 'image', 'label', 'link', + 'text') +metainfo = dict(classes=class_name) +_data_root = data_root + 'websiteScreenshots/' +dataset_websiteScreenshots = dict( + type=dataset_type, + metainfo=metainfo, + data_root=_data_root, + ann_file='valid/annotations_without_background.json', + data_prefix=dict(img='valid/'), + pipeline=_base_.test_pipeline, + test_mode=True, + return_classes=True) +val_evaluator_websiteScreenshots = dict( + type='CocoMetric', + ann_file=_data_root + 'valid/annotations_without_background.json', + metric='bbox') + +# --------------------- Config---------------------# +dataset_prefixes = [ + 'AerialMaritimeDrone_large', + 'AerialMaritimeDrone_tiled', + 'AmericanSignLanguageLetters', + 'Aquarium', + 'BCCD', + 'boggleBoards', + 'brackishUnderwater', + 'ChessPieces', + 'CottontailRabbits', + 'dice', + 'DroneControl', + 'EgoHands_generic', + 'EgoHands_specific', + 'HardHatWorkers', + 'MaskWearing', + 'MountainDewCommercial', + 'NorthAmericaMushrooms', + 'openPoetryVision', + 'OxfordPets_by_breed', + 'OxfordPets_by_species', + 'PKLot', + 'Packages', + 'PascalVOC', + 'pistols', + 'plantdoc', + 'pothole', + 'Raccoons', + 'selfdrivingCar', + 'ShellfishOpenImages', + 'ThermalCheetah', + 'thermalDogsAndPeople', + 'UnoCards', + 'VehiclesOpenImages', + 'WildfireSmoke', + 'websiteScreenshots', +] + +datasets = [ + dataset_AerialMaritimeDrone_large, dataset_AerialMaritimeDrone_tiled, + dataset_AmericanSignLanguageLetters, dataset_Aquarium, dataset_BCCD, + dataset_boggleBoards, dataset_brackishUnderwater, dataset_ChessPieces, + dataset_CottontailRabbits, dataset_dice, dataset_DroneControl, + dataset_EgoHands_generic, dataset_EgoHands_specific, + dataset_HardHatWorkers, dataset_MaskWearing, dataset_MountainDewCommercial, + dataset_NorthAmericaMushrooms, dataset_openPoetryVision, + dataset_OxfordPets_by_breed, dataset_OxfordPets_by_species, dataset_PKLot, + dataset_Packages, dataset_PascalVOC, dataset_pistols, dataset_plantdoc, + dataset_pothole, dataset_Raccoon, dataset_selfdrivingCar, + dataset_ShellfishOpenImages, dataset_ThermalCheetah, + dataset_thermalDogsAndPeople, dataset_UnoCards, dataset_VehiclesOpenImages, + dataset_WildfireSmoke, dataset_websiteScreenshots +] + +metrics = [ + val_evaluator_AerialMaritimeDrone_large, + val_evaluator_AerialMaritimeDrone_tiled, + val_evaluator_AmericanSignLanguageLetters, val_evaluator_Aquarium, + val_evaluator_BCCD, val_evaluator_boggleBoards, + val_evaluator_brackishUnderwater, val_evaluator_ChessPieces, + val_evaluator_CottontailRabbits, val_evaluator_dice, + val_evaluator_DroneControl, val_evaluator_EgoHands_generic, + val_evaluator_EgoHands_specific, val_evaluator_HardHatWorkers, + val_evaluator_MaskWearing, val_evaluator_MountainDewCommercial, + val_evaluator_NorthAmericaMushrooms, val_evaluator_openPoetryVision, + val_evaluator_OxfordPets_by_breed, val_evaluator_OxfordPets_by_species, + val_evaluator_PKLot, val_evaluator_Packages, val_evaluator_PascalVOC, + val_evaluator_pistols, val_evaluator_plantdoc, val_evaluator_pothole, + val_evaluator_Raccoon, val_evaluator_selfdrivingCar, + val_evaluator_ShellfishOpenImages, val_evaluator_ThermalCheetah, + val_evaluator_thermalDogsAndPeople, val_evaluator_UnoCards, + val_evaluator_VehiclesOpenImages, val_evaluator_WildfireSmoke, + val_evaluator_websiteScreenshots +] + +# -------------------------------------------------# +val_dataloader = dict( + dataset=dict(_delete_=True, type='ConcatDataset', datasets=datasets)) +test_dataloader = val_dataloader + +val_evaluator = dict( + _delete_=True, + type='MultiDatasetsEvaluator', + metrics=metrics, + dataset_prefixes=dataset_prefixes) +test_evaluator = val_evaluator diff --git a/configs/odinw/override_category.py b/configs/odinw/override_category.py new file mode 100644 index 00000000000..7641c837053 --- /dev/null +++ b/configs/odinw/override_category.py @@ -0,0 +1,96 @@ +# Copyright (c) OpenMMLab. All rights reserved. +import argparse + +import mmengine + + +def parse_args(): + parser = argparse.ArgumentParser(description='Override Category') + parser.add_argument('data_root') + return parser.parse_args() + + +def main(): + args = parse_args() + + ChessPieces = [{ + 'id': 1, + 'name': ' ', + 'supercategory': 'pieces' + }, { + 'id': 2, + 'name': 'black bishop', + 'supercategory': 'pieces' + }, { + 'id': 3, + 'name': 'black king', + 'supercategory': 'pieces' + }, { + 'id': 4, + 'name': 'black knight', + 'supercategory': 'pieces' + }, { + 'id': 5, + 'name': 'black pawn', + 'supercategory': 'pieces' + }, { + 'id': 6, + 'name': 'black queen', + 'supercategory': 'pieces' + }, { + 'id': 7, + 'name': 'black rook', + 'supercategory': 'pieces' + }, { + 'id': 8, + 'name': 'white bishop', + 'supercategory': 'pieces' + }, { + 'id': 9, + 'name': 'white king', + 'supercategory': 'pieces' + }, { + 'id': 10, + 'name': 'white knight', + 'supercategory': 'pieces' + }, { + 'id': 11, + 'name': 'white pawn', + 'supercategory': 'pieces' + }, { + 'id': 12, + 'name': 'white queen', + 'supercategory': 'pieces' + }, { + 'id': 13, + 'name': 'white rook', + 'supercategory': 'pieces' + }] + + _data_root = args.data_root + 'ChessPieces/Chess Pieces.v23-raw.coco/' + json_data = mmengine.load(_data_root + + 'valid/annotations_without_background.json') + json_data['categories'] = ChessPieces + mmengine.dump(json_data, + _data_root + 'valid/new_annotations_without_background.json') + + NorthAmericaMushrooms = [{ + 'id': 1, + 'name': 'flat mushroom', + 'supercategory': 'mushroom' + }, { + 'id': 2, + 'name': 'yellow mushroom', + 'supercategory': 'mushroom' + }] + + _data_root = args.data_root + 'NorthAmericaMushrooms/North American Mushrooms.v1-416x416.coco/' # noqa + json_data = mmengine.load(_data_root + + 'valid/annotations_without_background.json') + json_data['categories'] = NorthAmericaMushrooms + mmengine.dump(json_data, + _data_root + 'valid/new_annotations_without_background.json') + + +if __name__ == '__main__': + main() diff --git a/docs/zh_cn/user_guides/dataset_prepare.md b/docs/zh_cn/user_guides/dataset_prepare.md index a8bf32011a7..1caad856af0 100644 --- a/docs/zh_cn/user_guides/dataset_prepare.md +++ b/docs/zh_cn/user_guides/dataset_prepare.md @@ -305,3 +305,58 @@ mim download mmdet --dataset voc2012 # download coco2017 and preprocess by MIM mim download mmdet --dataset coco2017 ``` + +### ODinW 数据集准备 + +ODinW 数据集来自 GLIP 论文，用于评估预训练模型泛化性能。一共包括 ODinW-13 和 ODinW-35 两个版本，其中 ODinW-35 包括了 ODinW-13 的所有数据。 目前数据托管在 [huggingface](https://huggingface.co/GLIPModel/GLIP) + +请确保你提前安装好了 [git lfs](https://git-lfs.com), 然后按照如下命令下载 + +```shell +cd mmdetection + +git lfs install +# 我们不需要下载权重 +GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/GLIPModel/GLIP + +cd GLIP +git lfs pull --include="odinw_35" +``` + +下载完成后，目录结构如下所示： + +```text +mmdetection +├── GLIP +| ├── odinw_35 +| | ├── AerialMaritimeDrone.zip +| | ├── AmericanSignLanguageLetters.zip +... +``` + +你可以采用如下命令全部解压并移动到 `mmdetection/data` 路径下： + +```shell +#!/bin/bash + +folder="./GLIP/odinw_35/" + +find "$folder" -type f -name "*.zip" | while read -r file; do + unzip "$file" -d "$(dirname "$file")" +done + +mv GLIP/odinw_35 data/ +``` + +最终结构如下所示： + +```text +mmdetection +├── tools +├── configs +├── data +| ├── odinw_35 +| | ├── AerialMaritimeDrone +... +│ ├── coco +``` diff --git a/mmdet/datasets/base_det_dataset.py b/mmdet/datasets/base_det_dataset.py index 57bc7098387..8b3876d5c06 100644 --- a/mmdet/datasets/base_det_dataset.py +++ b/mmdet/datasets/base_det_dataset.py @@ -21,6 +21,8 @@ class BaseDetDataset(BaseDataset): corresponding backend. Defaults to None. return_classes (bool): Whether to return class information for open vocabulary-based algorithms. Defaults to False. + caption_prompt (dict, optional): Prompt for captioning. + Defaults to None. """ def __init__(self, @@ -30,11 +32,16 @@ def __init__(self, file_client_args: dict = None, backend_args: dict = None, return_classes: bool = False, + caption_prompt: Optional[dict] = None, **kwargs) -> None: self.seg_map_suffix = seg_map_suffix self.proposal_file = proposal_file self.backend_args = backend_args self.return_classes = return_classes + self.caption_prompt = caption_prompt + if self.caption_prompt is not None: + assert self.return_classes, \ + 'return_classes must be True when using caption_prompt' if file_client_args is not None: raise RuntimeError( 'The `file_client_args` is deprecated, ' diff --git a/mmdet/datasets/coco.py b/mmdet/datasets/coco.py index 277b75988da..1cf21c4e667 100644 --- a/mmdet/datasets/coco.py +++ b/mmdet/datasets/coco.py @@ -129,6 +129,7 @@ def parse_data_info(self, raw_data_info: dict) -> Union[dict, List[dict]]: if self.return_classes: data_info['text'] = self.metainfo['classes'] + data_info['caption_prompt'] = self.caption_prompt data_info['custom_entities'] = True instances = [] diff --git a/mmdet/datasets/dataset_wrappers.py b/mmdet/datasets/dataset_wrappers.py index e651e2b9902..d4e26e07c0f 100644 --- a/mmdet/datasets/dataset_wrappers.py +++ b/mmdet/datasets/dataset_wrappers.py @@ -247,6 +247,14 @@ def __init__(self, if not lazy_init: self.full_init() + if is_all_same: + self._metainfo.update( + dict(cumulative_sizes=self.cumulative_sizes)) + else: + for i, dataset in enumerate(self.datasets): + self._metainfo[i].update( + dict(cumulative_sizes=self.cumulative_sizes)) + def get_dataset_source(self, idx: int) -> int: dataset_idx, _ = self._get_ori_dataset_idx(idx) return dataset_idx diff --git a/mmdet/evaluation/__init__.py b/mmdet/evaluation/__init__.py index f70dc226d30..126dea092eb 100644 --- a/mmdet/evaluation/__init__.py +++ b/mmdet/evaluation/__init__.py @@ -1,3 +1,4 @@ # Copyright (c) OpenMMLab. All rights reserved. +from .evaluator import * # noqa: F401,F403 from .functional import * # noqa: F401,F403 from .metrics import * # noqa: F401,F403 diff --git a/mmdet/evaluation/evaluator/__init__.py b/mmdet/evaluation/evaluator/__init__.py new file mode 100644 index 00000000000..6b13fe99548 --- /dev/null +++ b/mmdet/evaluation/evaluator/__init__.py @@ -0,0 +1,4 @@ +# Copyright (c) OpenMMLab. All rights reserved. +from .multi_datasets_evaluator import MultiDatasetsEvaluator + +__all__ = ['MultiDatasetsEvaluator'] diff --git a/mmdet/evaluation/evaluator/multi_datasets_evaluator.py b/mmdet/evaluation/evaluator/multi_datasets_evaluator.py new file mode 100644 index 00000000000..5cff1cf210e --- /dev/null +++ b/mmdet/evaluation/evaluator/multi_datasets_evaluator.py @@ -0,0 +1,111 @@ +# Copyright (c) OpenMMLab. All rights reserved. +import warnings +from collections import OrderedDict +from typing import Sequence, Union + +from mmengine.dist import (broadcast_object_list, collect_results, + is_main_process) +from mmengine.evaluator import BaseMetric, Evaluator +from mmengine.evaluator.metric import _to_cpu +from mmengine.registry import EVALUATOR + +from mmdet.utils import ConfigType + + +@EVALUATOR.register_module() +class MultiDatasetsEvaluator(Evaluator): + """Wrapper class to compose class: `ConcatDataset` and multiple + :class:`BaseMetric` instances. + The metrics will be evaluated on each dataset slice separately. The name of + the each metric is the concatenation of the dataset prefix, the metric + prefix and the key of metric - e.g. + `dataset_prefix/metric_prefix/accuracy`. + + Args: + metrics (dict or BaseMetric or Sequence): The config of metrics. + dataset_prefixes (Sequence[str]): The prefix of each dataset. The + length of this sequence should be the same as the length of the + datasets. + """ + + def __init__(self, metrics: Union[ConfigType, BaseMetric, Sequence], + dataset_prefixes: Sequence[str]) -> None: + super().__init__(metrics) + self.dataset_prefixes = dataset_prefixes + self._setups = False + + def _get_cumulative_sizes(self): + # ConcatDataset have a property `cumulative_sizes` + if isinstance(self.dataset_meta, Sequence): + dataset_slices = self.dataset_meta[0]['cumulative_sizes'] + if not self._setups: + self._setups = True + for dataset_meta, metric in zip(self.dataset_meta, + self.metrics): + metric.dataset_meta = dataset_meta + else: + dataset_slices = self.dataset_meta['cumulative_sizes'] + return dataset_slices + + def evaluate(self, size: int) -> dict: + """Invoke ``evaluate`` method of each metric and collect the metrics + dictionary. + + Args: + size (int): Length of the entire validation dataset. When batch + size > 1, the dataloader may pad some data samples to make + sure all ranks have the same length of dataset slice. The + ``collect_results`` function will drop the padded data based on + this size. + + Returns: + dict: Evaluation results of all metrics. The keys are the names + of the metrics, and the values are corresponding results. + """ + metrics_results = OrderedDict() + dataset_slices = self._get_cumulative_sizes() + assert len(dataset_slices) == len(self.dataset_prefixes) + + for dataset_prefix, start, end, metric in zip( + self.dataset_prefixes, [0] + dataset_slices[:-1], + dataset_slices, self.metrics): + if len(metric.results) == 0: + warnings.warn( + f'{metric.__class__.__name__} got empty `self.results`.' + 'Please ensure that the processed results are properly ' + 'added into `self.results` in `process` method.') + + results = collect_results(metric.results, size, + metric.collect_device) + + if is_main_process(): + # cast all tensors in results list to cpu + results = _to_cpu(results) + _metrics = metric.compute_metrics( + results[start:end]) # type: ignore + + if metric.prefix: + final_prefix = '/'.join((dataset_prefix, metric.prefix)) + else: + final_prefix = dataset_prefix + print(f'================{final_prefix}================') + metric_results = { + '/'.join((final_prefix, k)): v + for k, v in _metrics.items() + } + + # Check metric name conflicts + for name in metric_results.keys(): + if name in metrics_results: + raise ValueError( + 'There are multiple evaluation results with ' + f'the same metric name {name}. Please make ' + 'sure all metrics have different prefixes.') + metrics_results.update(metric_results) + metric.results.clear() + if is_main_process(): + metrics_results = [metrics_results] + else: + metrics_results = [None] # type: ignore + broadcast_object_list(metrics_results) + return metrics_results[0] diff --git a/mmdet/models/detectors/glip.py b/mmdet/models/detectors/glip.py index e076a55fe20..13cfea960a8 100644 --- a/mmdet/models/detectors/glip.py +++ b/mmdet/models/detectors/glip.py @@ -1,7 +1,7 @@ # Copyright (c) OpenMMLab. All rights reserved. import re import warnings -from typing import Tuple, Union +from typing import Optional, Tuple, Union import torch from torch import Tensor @@ -208,9 +208,11 @@ def __init__(self, self._special_tokens = '. ' def get_tokens_and_prompts( - self, - original_caption: Union[str, list, tuple], - custom_entities: bool = False) -> Tuple[dict, str, list, list]: + self, + original_caption: Union[str, list, tuple], + custom_entities: bool = False, + enhanced_text_prompts: Optional[ConfigType] = None + ) -> Tuple[dict, str, list, list]: """Get the tokens positive and prompts for the caption.""" if isinstance(original_caption, (list, tuple)) or custom_entities: if custom_entities and isinstance(original_caption, str): @@ -219,15 +221,45 @@ def get_tokens_and_prompts( original_caption = list( filter(lambda x: len(x) > 0, original_caption)) - caption_string = '' - tokens_positive = [] - for idx, word in enumerate(original_caption): - tokens_positive.append( - [[len(caption_string), - len(caption_string) + len(word)]]) - caption_string += word - if idx != len(original_caption) - 1: - caption_string += self._special_tokens + if custom_entities and enhanced_text_prompts is not None: + caption_string = '' + tokens_positive = [] + for idx, word in enumerate(original_caption): + if word in enhanced_text_prompts: + enhanced_text_dict = enhanced_text_prompts[word] + if 'prefix' in enhanced_text_dict: + caption_string += enhanced_text_dict['prefix'] + start_i = len(caption_string) + if 'name' in enhanced_text_dict: + caption_string += enhanced_text_dict['name'] + else: + caption_string += word + end_i = len(caption_string) + tokens_positive.append([[start_i, end_i]]) + + if 'suffix' in enhanced_text_dict: + caption_string += enhanced_text_dict['suffix'] + else: + tokens_positive.append([[ + len(caption_string), + len(caption_string) + len(word) + ]]) + caption_string += word + + if idx != len(original_caption) - 1: + caption_string += self._special_tokens + else: + caption_string = '' + tokens_positive = [] + for idx, word in enumerate(original_caption): + tokens_positive.append([[ + len(caption_string), + len(caption_string) + len(word) + ]]) + caption_string += word + if idx != len(original_caption) - 1: + caption_string += self._special_tokens + tokenized = self.language_model.tokenizer([caption_string], return_tensors='pt') entities = original_caption @@ -248,12 +280,14 @@ def get_positive_map(self, tokenized, tokens_positive): return positive_map_label_to_token, positive_map def get_tokens_positive_and_prompts( - self, - original_caption: Union[str, list, tuple], - custom_entities: bool = False) -> Tuple[dict, str, Tensor, list]: + self, + original_caption: Union[str, list, tuple], + custom_entities: bool = False, + enhanced_text_prompt: Optional[ConfigType] = None + ) -> Tuple[dict, str, Tensor, list]: tokenized, caption_string, tokens_positive, entities = \ self.get_tokens_and_prompts( - original_caption, custom_entities) + original_caption, custom_entities, enhanced_text_prompt) positive_map_label_to_token, positive_map = self.get_positive_map( tokenized, tokens_positive) return positive_map_label_to_token, caption_string, \ @@ -345,6 +379,9 @@ def predict(self, text_prompts = [ data_samples.text for data_samples in batch_data_samples ] + enhanced_text_prompts = [ + data_samples.caption_prompt for data_samples in batch_data_samples + ] if 'custom_entities' in batch_data_samples[0]: # Assuming that the `custom_entities` flag @@ -357,14 +394,16 @@ def predict(self, # All the text prompts are the same, # so there is no need to calculate them multiple times. _positive_maps_and_prompts = [ - self.get_tokens_positive_and_prompts(text_prompts[0], - custom_entities) + self.get_tokens_positive_and_prompts( + text_prompts[0], custom_entities, enhanced_text_prompts[0]) ] * len(batch_inputs) else: _positive_maps_and_prompts = [ self.get_tokens_positive_and_prompts(text_prompt, - custom_entities) - for text_prompt in text_prompts + custom_entities, + enhanced_text_prompt) + for text_prompt, enhanced_text_prompt in zip( + text_prompts, enhanced_text_prompts) ] token_positive_maps, text_prompts, _, entities = zip(