[Refactor] Support to use "split" to specify training set/validation set in the ImageNet dataset (#1535)

* [Feature]: Add caption

* [Feature]: Update scienceqa

* [CI] Add test mim CI. (#879)

* refactor imagenet dataset

* refactor imagenet dataset

* refactor imagenet dataset

* update imagenet21k

* update configs

* update mnist

* update dataset_prepare.md

* fix sun397 url and update user_guides/dataset_prepare.md

* update dataset_prepare.md

* fix sun397 dataset

* fix sun397

* update chinese dataset_prepare.md

* update dataset_prepare.md

* [Refactor] update voc dataset

* [Refactor] update voc dataset

* refactor imagenet

* refactor imagenet

* use mmengine.fileio

---------

Co-authored-by: liuyuan <3463423099@qq.com>
Co-authored-by: Ma Zerun <mzr1996@163.com>
Co-authored-by: Ezra-Yu <18586273+Ezra-Yu@users.noreply.github.com>

configs/_base_/datasets/imagenet21k_bs128.py

@@ -16,38 +16,13 @@ train_pipeline = [
     dict(type='PackInputs'),
 ]
 
-test_pipeline = [
-    dict(type='LoadImageFromFile'),
-    dict(type='ResizeEdge', scale=256, edge='short'),
-    dict(type='CenterCrop', crop_size=224),
-    dict(type='PackInputs'),
-]
-
 train_dataloader = dict(
     batch_size=128,
     num_workers=5,
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet21k',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
-
-val_dataloader = dict(
-    batch_size=128,
-    num_workers=5,
-    dataset=dict(
-        type=dataset_type,
-        data_root='data/imagenet21k',
-        ann_file='meta/val.txt',
-        data_prefix='val',
-        pipeline=test_pipeline),
-    sampler=dict(type='DefaultSampler', shuffle=False),
-)
-val_evaluator = dict(type='Accuracy', topk=(1, 5))
-
-# If you want standard test, please manually configure the test dataset
-test_dataloader = val_dataloader
-test_evaluator = val_evaluator

configs/_base_/datasets/imagenet_bs128_mbv3.py

@@ -44,8 +44,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -56,8 +55,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs128_poolformer_medium_224.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs128_poolformer_small_224.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs128_revvit_224.py

@@ -59,8 +59,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
     persistent_workers=True,
@@ -72,8 +71,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
     persistent_workers=True,

configs/_base_/datasets/imagenet_bs128_riformer_medium_384.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs128_riformer_small_384.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs128_vig_224.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs16_eva_196.py

@@ -38,8 +38,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -50,8 +49,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs16_eva_336.py

@@ -38,8 +38,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -50,8 +49,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs16_eva_560.py

@@ -38,8 +38,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -50,8 +49,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs16_pil_bicubic_384.py

@@ -31,8 +31,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -43,8 +42,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs256_beitv2.py

@@ -43,6 +43,5 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix=dict(img_path='train/'),
         pipeline=train_pipeline))

configs/_base_/datasets/imagenet_bs256_davit_224.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs256_levit_224.py

@@ -57,9 +57,8 @@ train_dataloader = dict(
     num_workers=4,
     dataset=dict(
         type=dataset_type,
-        data_root=r'E:\imagenet',
+        data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='train',
-        data_prefix='ILSVRC2012_img_val',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -69,9 +68,8 @@ val_dataloader = dict(
     num_workers=4,
     dataset=dict(
         type=dataset_type,
-        data_root=r'E:\imagenet',
+        data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='ILSVRC2012_img_val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs256_rsb_a12.py

@@ -50,8 +50,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -62,8 +61,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs256_rsb_a3.py

@@ -50,8 +50,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -62,8 +61,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs256_simmim_192.py

@@ -29,6 +29,5 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix=dict(img_path='train/'),
         pipeline=train_pipeline))

configs/_base_/datasets/imagenet_bs256_swin_192.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
 )
 
@@ -72,8 +71,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
 )
 val_evaluator = dict(type='Accuracy', topk=(1, 5))

configs/_base_/datasets/imagenet_bs32.py

@@ -29,8 +29,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -41,8 +40,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs32_byol.py

@@ -85,6 +85,5 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix=dict(img_path='train/'),
         pipeline=train_pipeline))

configs/_base_/datasets/imagenet_bs32_mocov2.py

@@ -54,6 +54,5 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix=dict(img_path='train/'),
         pipeline=train_pipeline))

configs/_base_/datasets/imagenet_bs32_pil_bicubic.py

@@ -38,8 +38,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -50,8 +49,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs32_pil_resize.py

@@ -29,8 +29,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -41,8 +40,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs32_simclr.py

@@ -48,6 +48,5 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix=dict(img_path='train/'),
         pipeline=train_pipeline))

configs/_base_/datasets/imagenet_bs512_mae.py

@@ -28,6 +28,5 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix=dict(img_path='train/'),
         pipeline=train_pipeline))

configs/_base_/datasets/imagenet_bs512_mocov3.py

@@ -86,6 +86,5 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root=data_root,
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix=dict(img_path='train/'),
         pipeline=train_pipeline))

configs/_base_/datasets/imagenet_bs64.py

@@ -29,8 +29,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -41,8 +40,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_autoaug.py

@@ -37,8 +37,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -49,8 +48,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_clip_224.py

@@ -55,18 +55,19 @@ data = dict(
     workers_per_gpu=8,
     train=dict(
         type=dataset_type,
-        data_prefix='data/imagenet/train',
+        data_root='data/imagenet',
+        split='train',
         pipeline=train_pipeline),
     val=dict(
         type=dataset_type,
-        data_prefix='data/imagenet/val',
+        data_root='data/imagenet',
-        ann_file='data/imagenet/meta/val.txt',
+        split='val',
         pipeline=test_pipeline),
     test=dict(
         # replace `data/val` with `data/test` for standard test
         type=dataset_type,
-        data_prefix='data/imagenet/val',
+        data_root='data/imagenet',
-        ann_file='data/imagenet/meta/val.txt',
+        split='val',
         pipeline=test_pipeline))
 
 evaluation = dict(interval=10, metric='accuracy')

configs/_base_/datasets/imagenet_bs64_clip_384.py

@@ -55,18 +55,19 @@ data = dict(
     workers_per_gpu=8,
     train=dict(
         type=dataset_type,
-        data_prefix='data/imagenet/train',
+        data_root='data/imagenet',
+        split='train',
         pipeline=train_pipeline),
     val=dict(
         type=dataset_type,
-        data_prefix='data/imagenet/val',
+        data_root='data/imagenet',
-        ann_file='data/imagenet/meta/val.txt',
+        split='val',
         pipeline=test_pipeline),
     test=dict(
         # replace `data/val` with `data/test` for standard test
         type=dataset_type,
-        data_prefix='data/imagenet/val',
+        data_root='data/imagenet',
-        ann_file='data/imagenet/meta/val.txt',
+        split='val',
         pipeline=test_pipeline))
 
 evaluation = dict(interval=10, metric='accuracy')

configs/_base_/datasets/imagenet_bs64_clip_448.py

@@ -56,18 +56,19 @@ data = dict(
     workers_per_gpu=8,
     train=dict(
         type=dataset_type,
-        data_prefix='data/imagenet/train',
+        data_root='data/imagenet',
+        split='train',
         pipeline=train_pipeline),
     val=dict(
         type=dataset_type,
-        data_prefix='data/imagenet/val',
+        data_root='data/imagenet',
-        ann_file='data/imagenet/meta/val.txt',
+        split='val',
         pipeline=test_pipeline),
     test=dict(
         # replace `data/val` with `data/test` for standard test
         type=dataset_type,
-        data_prefix='data/imagenet/val',
+        data_root='data/imagenet',
-        ann_file='data/imagenet/meta/val.txt',
+        split='val',
         pipeline=test_pipeline))
 
 evaluation = dict(interval=10, metric='accuracy')

configs/_base_/datasets/imagenet_bs64_convmixer_224.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_deit3_224.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_deit3_384.py

@@ -38,8 +38,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -50,8 +49,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_edgenext_256.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_mixer_224.py

@@ -30,8 +30,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -42,8 +41,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_pil_resize.py

@@ -29,8 +29,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -41,8 +40,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_pil_resize_autoaug.py

@@ -46,8 +46,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -58,8 +57,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_swin_224.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_swin_256.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_swin_384.py

@@ -32,8 +32,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -44,8 +43,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs64_t2t_224.py

@@ -58,8 +58,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -70,8 +69,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/imagenet_bs8_pil_bicubic_320.py

@@ -37,8 +37,7 @@ train_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/train.txt',
+        split='train',
-        data_prefix='train',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -49,8 +48,7 @@ val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
         data_root='data/imagenet',
-        ann_file='meta/val.txt',
+        split='val',
-        data_prefix='val',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )

configs/_base_/datasets/voc_bs16.py

@@ -22,7 +22,12 @@ test_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(type='ResizeEdge', scale=256, edge='short'),
     dict(type='CenterCrop', crop_size=224),
-    dict(type='PackInputs'),
+    dict(
+        type='PackInputs',
+        # `gt_label_difficult` is needed for VOC evaluation
+        meta_keys=('sample_idx', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor', 'flip', 'flip_direction',
+                   'gt_label_difficult')),
 ]
 
 train_dataloader = dict(
@@ -30,8 +35,8 @@ train_dataloader = dict(
     num_workers=5,
     dataset=dict(
         type=dataset_type,
-        data_root='data/VOCdevkit/VOC2007',
+        data_root='data/VOC2007',
-        image_set_path='ImageSets/Layout/val.txt',
+        split='trainval',
         pipeline=train_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=True),
 )
@@ -41,22 +46,13 @@ val_dataloader = dict(
     num_workers=5,
     dataset=dict(
         type=dataset_type,
-        data_root='data/VOCdevkit/VOC2007',
+        data_root='data/VOC2007',
-        image_set_path='ImageSets/Layout/val.txt',
+        split='test',
         pipeline=test_pipeline),
     sampler=dict(type='DefaultSampler', shuffle=False),
 )
 
-test_dataloader = dict(
+test_dataloader = val_dataloader
-    batch_size=16,
-    num_workers=5,
-    dataset=dict(
-        type=dataset_type,
-        data_root='data/VOCdevkit/VOC2007',
-        image_set_path='ImageSets/Layout/val.txt',
-        pipeline=test_pipeline),
-    sampler=dict(type='DefaultSampler', shuffle=False),
-)
 
 # calculate precision_recall_f1 and mAP
 val_evaluator = [
@@ -65,6 +61,5 @@ val_evaluator = [
     dict(type='VOCAveragePrecision')
 ]
 
-# If you want standard test, please manually configure the test dataset
 test_dataloader = val_dataloader
 test_evaluator = val_evaluator

docs/en/user_guides/dataset_prepare.md

@@ -145,16 +145,14 @@ ImageNet has multiple versions, but the most commonly used one is [ILSVRC 2012](
    - ILSVRC2012_img_train.tar (~138GB)
    - ILSVRC2012_img_val.tar (~6.3GB)
 3. Untar the downloaded files
-4. Re-organize the image files according the format convention of [CustomDataset](#CustomDataset)
 
-```{note}
+### The Directory Structrue of the ImageNet dataset
-In MMPretrain, we use the text annotation file format ImageNet in all provided config files. Therefore, to use
-the subfolder format, you please set `ann_file=''` in these config files.
-```
 
-### Subfolder Format
+We support two ways of organizing the ImageNet dataset: Subfolder Format and Text Annotation File Format.
 
-Re-organize the dataset as below:
+#### Subfolder Format
+
+We have provided a sample, which you can download and extract from this [link](https://download.openmmlab.com/mmpretrain/datasets/imagenet_1k.zip). The directory structure of the dataset should be as below:
 
 ```text
 data/imagenet/
@@ -177,37 +175,7 @@ data/imagenet/
 │   ├── ...
 ```
 
-And then, you can use the [`ImageNet`](mmpretrain.datasets.ImageNet) dataset with the below configurations:
+#### Text Annotation File Format
-
-```python
-train_dataloader = dict(
-    ...
-    # Training dataset configurations
-    dataset=dict(
-        type='ImageNet',
-        data_root='data/imagenet',
-        data_prefix='train',
-        ann_file='',
-        pipeline=...,
-    )
-)
-
-val_dataloader = dict(
-    ...
-    # Validation dataset configurations
-    dataset=dict(
-        type='ImageNet',
-        data_root='data/imagenet',
-        data_prefix='val',
-        ann_file='',
-        pipeline=...,
-    )
-)
-
-test_dataloader = val_dataloader
-```
-
-### Text Annotation File Format
 
 You can download and untar the meta data from this [link](https://download.openmmlab.com/mmclassification/datasets/imagenet/meta/caffe_ilsvrc12.tar.gz). And re-organize the dataset as below:
 
@@ -235,7 +203,9 @@ data/imagenet/
 │   ├── ...
 ```
 
-And then, you can use the [`ImageNet`](mmpretrain.datasets.ImageNet) dataset with the below configurations:
+### Configuration
+
+Once your dataset is organized in the way described above, you can use the [`ImageNet`](mmpretrain.datasets.ImageNet) dataset with the below configurations:
 
 ```python
 train_dataloader = dict(
@@ -243,9 +213,8 @@ train_dataloader = dict(
     # Training dataset configurations
     dataset=dict(
         type='ImageNet',
-        data_root='imagenet_folder',
+        data_root='data/imagenet',
-        data_prefix='train/',
+        split='train',
-        ann_file='meta/train.txt',
         pipeline=...,
     )
 )
@@ -255,9 +224,8 @@ val_dataloader = dict(
     # Validation dataset configurations
     dataset=dict(
         type='ImageNet',
-        data_root='imagenet_folder',
+        data_root='data/imagenet',
-        data_prefix='val/',
+        split='val',
-        ann_file='meta/val.txt',
         pipeline=...,
     )
 )
@@ -265,69 +233,27 @@ val_dataloader = dict(
 test_dataloader = val_dataloader
 ```
 
-## CIFAR
+## Supported Image Classification Datasets
 
-We support downloading the [`CIFAR10`](mmpretrain.datasets.CIFAR10) and [`CIFAR100`](mmpretrain.datasets.CIFAR100) datasets automatically, and you just need to specify the
+| Datasets                                                                           | split                               | HomePage                                                                            |
-download folder in the `data_root` field. And please specify `test_mode=False` / `test_mode=True`
+| ---------------------------------------------------------------------------------- | :---------------------------------- | ----------------------------------------------------------------------------------- |
-to use training datasets or test datasets.
+| [`Calthch101`](mmpretrain.datasets.Caltech101)(data_root[, split, pipeline, ...])  | ["train", "test"]                   | [Caltech 101](https://data.caltech.edu/records/mzrjq-6wc02) Dataset.                |
+| [`CIFAR10`](mmpretrain.datasets.CIFAR10)(data_root[, split, pipeline, ...])        | ["train", "test"]                   | [CIFAR10](https://www.cs.toronto.edu/~kriz/cifar.html) Dataset.                     |
+| [`CIFAR100`](mmpretrain.datasets.CIFAR100)(data_root[, split, pipeline, ...])      | ["train", "test"]                   | [CIFAR100](https://www.cs.toronto.edu/~kriz/cifar.html) Dataset.                    |
+| [`CUB`](mmpretrain.datasets.CUB)(data_root[, split, pipeline, ...])                | ["train", "test"]                   | [CUB-200-2011](http://www.vision.caltech.edu/datasets/cub_200_2011/) Dataset.       |
+| [`DTD`](mmpretrain.datasets.DTD)(data_root[, split, pipeline, ...])                | ["train", "val", "tranval", "test"] | [Describable Texture Dataset (DTD)](https://www.robots.ox.ac.uk/~vgg/data/dtd/) Dataset. |
+| [`FashionMNIST`](mmpretrain.datasets.FashionMNIST) (data_root[, split, pipeline, ...]) | ["train", "test"]                   | [Fashion-MNIST](https://github.com/zalandoresearch/fashion-mnist) Dataset.          |
+| [`FGVCAircraft`](mmpretrain.datasets.FGVCAircraft)(data_root[, split, pipeline, ...]) | ["train", "val", "tranval", "test"] | [FGVC Aircraft](https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/) Dataset.      |
+| [`Flowers102`](mmpretrain.datasets.Flowers102)(data_root[, split, pipeline, ...])  | ["train", "val", "tranval", "test"] | [Oxford 102 Flower](https://www.robots.ox.ac.uk/~vgg/data/flowers/102/) Dataset.    |
+| [`Food101`](mmpretrain.datasets.Food101)(data_root[, split, pipeline, ...])        | ["train", "test"]                   | [Food101](https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/) Dataset.     |
+| [`MNIST`](mmpretrain.datasets.MNIST) (data_root[, split, pipeline, ...])           | ["train", "test"]                   | [MNIST](http://yann.lecun.com/exdb/mnist/) Dataset.                                 |
+| [`OxfordIIITPet`](mmpretrain.datasets.OxfordIIITPet)(data_root[, split, pipeline, ...]) | ["tranval", test"]                  | [Oxford-IIIT Pets](https://www.robots.ox.ac.uk/~vgg/data/pets/) Dataset.            |
+| [`Places205`](mmpretrain.datasets.Places205)(data_root[, pipeline, ...])           | -                                   | [Places205](http://places.csail.mit.edu/downloadData.html) Dataset.                 |
+| [`StanfordCars`](mmpretrain.datasets.StanfordCars)(data_root[, split, pipeline, ...]) | ["train", "test"]                   | [Stanford Cars](https://ai.stanford.edu/~jkrause/cars/car_dataset.html) Dataset.    |
+| [`SUN397`](mmpretrain.datasets.SUN397)(data_root[, split, pipeline, ...])          | ["train", "test"]                   | [SUN397](https://vision.princeton.edu/projects/2010/SUN/) Dataset.                  |
+| [`VOC`](mmpretrain.datasets.VOC)(data_root[, image_set_path, pipeline, ...])       | ["train", "val", "tranval", "test"] | [Pascal VOC](http://host.robots.ox.ac.uk/pascal/VOC/) Dataset.                      |
 
-```python
+Some dataset homepage links may be unavailable, and you can download datasets through [OpenDataLab](https://opendatalab.com/), such as [Stanford Cars](https://opendatalab.com/Stanford_Cars/download).
-train_dataloader = dict(
-    ...
-    # Training dataset configurations
-    dataset=dict(
-        type='CIFAR10',
-        data_root='data/cifar10',
-        test_mode=False,
-        pipeline=...,
-    )
-)
-
-val_dataloader = dict(
-    ...
-    # Validation dataset configurations
-    dataset=dict(
-        type='CIFAR10',
-        data_root='data/cifar10',
-        test_mode=True,
-        pipeline=...,
-    )
-)
-
-test_dataloader = val_dataloader
-```
-
-## MNIST
-
-We support downloading the [MNIST](mmpretrain.datasets.MNIST) and [Fashion-MNIST](mmpretrain.datasets.FashionMNIST) datasets automatically, and you just need to specify the
-download folder in the `data_root` field. And please specify `test_mode=False` / `test_mode=True`
-to use training datasets or test datasets.
-
-```python
-train_dataloader = dict(
-    ...
-    # Training dataset configurations
-    dataset=dict(
-        type='MNIST',
-        data_root='data/mnist',
-        test_mode=False,
-        pipeline=...,
-    )
-)
-
-val_dataloader = dict(
-    ...
-    # Validation dataset configurations
-    dataset=dict(
-        type='MNIST',
-        data_root='data/mnist',
-        test_mode=True,
-        pipeline=...,
-    )
-)
-
-test_dataloader = val_dataloader
-```
 
 ## OpenMMLab 2.0 Standard Dataset
 

docs/zh_CN/user_guides/dataset_prepare.md

@@ -143,15 +143,14 @@ ImageNet 有多个版本，但最常用的一个是 [ILSVRC 2012](http://www.ima
    - ILSVRC2012_img_train.tar (~138GB)
    - ILSVRC2012_img_val.tar (~6.3GB)
 3. 解压已下载的图片。
-4. 根据 [CustomDataset](#CustomDataset) 的格式约定重新组织图像文件
 
-```{note}
+### ImageNet数据集目录结构
-在 MMPretrain 中，所有配置文件默认使用文本标注文件的数据集。因此，如果希望使用子文件夹格式，请在配置文件中设置 `ann_file=''`。
-```
 
-### 子文件夹格式
+我们支持两种方式组织ImageNet数据集，子目录格式和文本注释文件格式。
 
-重新组织数据集如下：
+#### 子文件夹格式
+
+我们提供了一个样例，您可以从这个[链接](https://download.openmmlab.com/mmpretrain/datasets/imagenet_1k.zip)下载和解压。数据集的目录结构应如下所示：
 
 ```text
 data/imagenet/
@@ -174,37 +173,7 @@ data/imagenet/
 │   ├── ...
 ```
 
-然后，您可以使用具有以下配置的 [`ImageNet`](mmpretrain.datasets.ImageNet) 数据集：
+#### 文本标注文件格式
-
-```python
-train_dataloader = dict(
-    ...
-    # 训练数据集配置
-    dataset=dict(
-        type='ImageNet',
-        data_root='data/imagenet',
-        data_prefix='train',
-        ann_file='',
-        pipeline=...,
-    )
-)
-
-val_dataloader = dict(
-    ...
-    # 验证数据集配置
-    dataset=dict(
-        type='ImageNet',
-        data_root='data/imagenet',
-        data_prefix='val',
-        ann_file='',
-        pipeline=...,
-    )
-)
-
-test_dataloader = val_dataloader
-```
-
-### 文本标注文件格式
 
 您可以从[此链接](https://download.openmmlab.com/mmclassification/datasets/imagenet/meta/caffe_ilsvrc12.tar.gz)下载并解压元数据，然后组织文件夹如下：
 
@@ -232,7 +201,9 @@ data/imagenet/
 │   ├── ...
 ```
 
-然后，您可以使用具有以下配置的 [`ImageNet`](mmpretrain.datasets.ImageNet) 数据集：
+### 配置
+
+当您的数据集以上述方式组织时，您可以使用具有以下配置的 [`ImageNet`](mmpretrain.datasets.ImageNet) 数据集：
 
 ```python
 train_dataloader = dict(
@@ -240,9 +211,8 @@ train_dataloader = dict(
     # 训练数据集配置
     dataset=dict(
         type='ImageNet',
-        data_root='imagenet_folder',
+        data_root='data/imagenet/',
-        data_prefix='train/',
+        split='train',
-        ann_file='meta/train.txt',
         pipeline=...,
     )
 )
@@ -252,9 +222,8 @@ val_dataloader = dict(
     # 验证数据集配置
     dataset=dict(
         type='ImageNet',
-        data_root='imagenet_folder',
+        data_root='data/imagenet/',
-        data_prefix='val/',
+        split='val',
-        ann_file='meta/val.txt',
         pipeline=...,
     )
 )
@@ -262,61 +231,27 @@ val_dataloader = dict(
 test_dataloader = val_dataloader
 ```
 
-## CIFAR
+## 支持的图像分类数据集
 
-我们支持自动下载 [`CIFAR10`](mmpretrain.datasets.CIFAR10) 和 [`CIFAR100`](mmpretrain.datasets.CIFAR100) 数据集，您只需在 `data_root` 字段中指定下载文件夹即可。 并且通过指定 `test_mode=False` / `test_mode=True` 来使用训练数据集或测试数据集。
+| 数据集                                                                              | split                               | 主页                                                                               |
+| ----------------------------------------------------------------------------------- | ----------------------------------- | ---------------------------------------------------------------------------------- |
+| [`Calthch101`](mmpretrain.datasets.Caltech101)(data_root[, split, pipeline, ...])   | ["train", "test"]                   | [Caltech 101](https://data.caltech.edu/records/mzrjq-6wc02) 数据集                 |
+| [`CIFAR10`](mmpretrain.datasets.CIFAR10)(data_root[, split, pipeline, ...])         | ["train", "test"]                   | [CIFAR10](https://www.cs.toronto.edu/~kriz/cifar.html) 数据集                      |
+| [`CIFAR100`](mmpretrain.datasets.CIFAR100)(data_root[, split, pipeline, ...])       | ["train", "test"]                   | [CIFAR100](https://www.cs.toronto.edu/~kriz/cifar.html) 数据集                     |
+| [`CUB`](mmpretrain.datasets.CUB)(data_root[, split, pipeline, ...])                 | ["train", "test"]                   | [CUB-200-2011](http://www.vision.caltech.edu/datasets/cub_200_2011/) 数据集        |
+| [`DTD`](mmpretrain.datasets.DTD)(data_root[, split, pipeline, ...])                 | ["train", "val", "tranval", "test"] | [Describable Texture Dataset (DTD)](https://www.robots.ox.ac.uk/~vgg/data/dtd/) 数据集 |
+| [`FashionMNIST`](mmpretrain.datasets.FashionMNIST) (data_root[, split, pipeline, ...]) | ["train", "test"]                   | [Fashion-MNIST](https://github.com/zalandoresearch/fashion-mnist) 数据集           |
+| [`FGVCAircraft`](mmpretrain.datasets.FGVCAircraft)(data_root[, split, pipeline, ...]) | ["train", "val", "tranval", "test"] | [FGVC Aircraft](https://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/) 数据集       |
+| [`Flowers102`](mmpretrain.datasets.Flowers102)(data_root[, split, pipeline, ...])   | ["train", "val", "tranval", "test"] | [Oxford 102 Flower](https://www.robots.ox.ac.uk/~vgg/data/flowers/102/) 数据集     |
+| [`Food101`](mmpretrain.datasets.Food101)(data_root[, split, pipeline, ...])         | ["train", "test"]                   | [Food101](https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/) 数据集      |
+| [`MNIST`](mmpretrain.datasets.MNIST) (data_root[, split, pipeline, ...])            | ["train", "test"]                   | [MNIST](http://yann.lecun.com/exdb/mnist/) 数据集                                  |
+| [`OxfordIIITPet`](mmpretrain.datasets.OxfordIIITPet)(data_root[, split, pipeline, ...]) | ["tranval", test"]                  | [Oxford-IIIT Pets](https://www.robots.ox.ac.uk/~vgg/data/pets/) 数据集             |
+| [`Places205`](mmpretrain.datasets.Places205)(data_root[, pipeline, ...])            | -                                   | [Places205](http://places.csail.mit.edu/downloadData.html) 数据集                  |
+| [`StanfordCars`](mmpretrain.datasets.StanfordCars)(data_root[, split, pipeline, ...]) | ["train", "test"]                   | [StanfordCars](https://ai.stanford.edu/~jkrause/cars/car_dataset.html) 数据集      |
+| [`SUN397`](mmpretrain.datasets.SUN397)(data_root[, split, pipeline, ...])           | ["train", "test"]                   | [SUN397](https://vision.princeton.edu/projects/2010/SUN/) 数据集                   |
+| [`VOC`](mmpretrain.datasets.VOC)(data_root[, image_set_path, pipeline, ...])        | ["train", "val", "tranval", "test"] | [Pascal VOC](http://host.robots.ox.ac.uk/pascal/VOC/) 数据集                       |
 
-```python
+有些数据集主页链接可能已经失效，您可以通过[OpenDataLab](https://opendatalab.com/)下载数据集，例如 [Stanford Cars](https://opendatalab.com/Stanford_Cars/download)数据集。
-train_dataloader = dict(
-    ...
-    # 训练数据集配置
-    dataset=dict(
-        type='CIFAR10',
-        data_root='data/cifar10',
-        test_mode=False,
-        pipeline=...,
-    )
-)
-val_dataloader = dict(
-    ...
-    # 验证数据集配置
-    dataset=dict(
-        type='CIFAR10',
-        data_root='data/cifar10',
-        test_mode=True,
-        pipeline=...,
-    )
-)
-test_dataloader = val_dataloader
-```
-
-## MNIST
-
-我们支持自动下载 [MNIST](mmpretrain.datasets.MNIST) 和 [Fashion-MNIST](mmpretrain.datasets.FashionMNIST) 数据集，您只需指定 `data_root` 字段中的下载路径即可。 并且通过指定 `test_mode=False` / `test_mode=True` 来使用训练数据集或测试数据集。
-
-```python
-train_dataloader = dict(
-    ...
-    # 训练数据集配置
-    dataset=dict(
-        type='MNIST',
-        data_root='data/mnist',
-        test_mode=False,
-        pipeline=...,
-    )
-)
-val_dataloader = dict(
-    ...
-    # 验证数据集配置
-    dataset=dict(
-        type='MNIST',
-        data_root='data/mnist',
-        test_mode=True,
-        pipeline=...,
-    )
-)
-test_dataloader = val_dataloader
-```
 
 ## OpenMMLab 2.0 标准数据集
 

mmpretrain/datasets/imagenet.py

@@ -1,6 +1,7 @@
 # Copyright (c) OpenMMLab. All rights reserved.
-from typing import Optional, Union
+from typing import List, Optional, Union
 
+from mmengine import fileio
 from mmengine.logging import MMLogger
 
 from mmpretrain.registry import DATASETS
@@ -12,18 +13,79 @@ from .custom import CustomDataset
 class ImageNet(CustomDataset):
     """`ImageNet <http://www.image-net.org>`_ Dataset.
 
-    The dataset supports two kinds of annotation format. More details can be
+    The dataset supports two kinds of directory format,
-    found in :class:`CustomDataset`.
+
+    ::
+
+        imagenet
+        ├── train
+        │   ├──class_x
+        |   |   ├── x1.jpg
+        |   |   ├── x2.jpg
+        |   |   └── ...
+        │   ├── class_y
+        |   |   ├── y1.jpg
+        |   |   ├── y2.jpg
+        |   |   └── ...
+        |   └── ...
+        ├── val
+        │   ├──class_x
+        |   |   └── ...
+        │   ├── class_y
+        |   |   └── ...
+        |   └── ...
+        └── test
+            ├── test1.jpg
+            ├── test2.jpg
+            └── ...
+
+    or ::
+
+        imagenet
+        ├── train
+        │   ├── x1.jpg
+        │   ├── y1.jpg
+        │   └── ...
+        ├── val
+        │   ├── x3.jpg
+        │   ├── y3.jpg
+        │   └── ...
+        ├── test
+        │   ├── test1.jpg
+        │   ├── test2.jpg
+        │   └── ...
+        └── meta
+            ├── train.txt
+            └── val.txt
+
 
     Args:
         data_root (str): The root directory for ``data_prefix`` and
             ``ann_file``. Defaults to ''.
+        split (str): The dataset split, supports "train", "val" and "test".
+            Default to ''.
         data_prefix (str | dict): Prefix for training data. Defaults to ''.
         ann_file (str): Annotation file path. Defaults to ''.
         metainfo (dict, optional): Meta information for dataset, such as class
             information. Defaults to None.
         **kwargs: Other keyword arguments in :class:`CustomDataset` and
             :class:`BaseDataset`.
+
+
+    Examples:
+        >>> from mmpretrain.datasets import ImageNet
+        >>> train_dataset = ImageNet(data_root='data/imagenet', split='train')
+        >>> train_dataset
+        Dataset ImageNet
+            Number of samples:  1281167
+            Number of categories:       1000
+            Root of dataset:    data/imagenet
+        >>> test_dataset = ImageNet(data_root='data/imagenet', split='val')
+        >>> test_dataset
+        Dataset ImageNet
+            Number of samples:  50000
+            Number of categories:       1000
+            Root of dataset:    data/imagenet
     """  # noqa: E501
 
     IMG_EXTENSIONS = ('.jpg', '.jpeg', '.png', '.ppm', '.bmp', '.pgm', '.tif')
@@ -31,11 +93,32 @@ class ImageNet(CustomDataset):
 
     def __init__(self,
                  data_root: str = '',
+                 split: str = '',
                  data_prefix: Union[str, dict] = '',
                  ann_file: str = '',
                  metainfo: Optional[dict] = None,
                  **kwargs):
         kwargs = {'extensions': self.IMG_EXTENSIONS, **kwargs}
+
+        if split:
+            splits = ['train', 'val', 'test']
+            assert split in splits, \
+                f"The split must be one of {splits}, but get '{split}'"
+
+            if split == 'test':
+                logger = MMLogger.get_current_instance()
+                logger.info(
+                    'Since the ImageNet1k test set does not provide label'
+                    'annotations, `with_label` is set to False')
+                kwargs['with_label'] = False
+
+            data_prefix = split if data_prefix == '' else data_prefix
+
+            if ann_file == '':
+                _ann_path = fileio.join_path(data_root, 'meta', f'{split}.txt')
+                if fileio.exists(_ann_path):
+                    ann_file = _ann_path
+
         super().__init__(
             data_root=data_root,
             data_prefix=data_prefix,
@@ -43,14 +126,39 @@ class ImageNet(CustomDataset):
             metainfo=metainfo,
             **kwargs)
 
+    def extra_repr(self) -> List[str]:
+        """The extra repr information of the dataset."""
+        body = [
+            f'Root of dataset: \t{self.data_root}',
+        ]
+        return body
+
 
 @DATASETS.register_module()
 class ImageNet21k(CustomDataset):
     """ImageNet21k Dataset.
 
-    Since the dataset ImageNet21k is extremely big, cantains 21k+ classes
+    Since the dataset ImageNet21k is extremely big, contains 21k+ classes
     and 1.4B files. We won't provide the default categories list. Please
     specify it from the ``classes`` argument.
+    The dataset directory structure is as follows,
+
+    ImageNet21k dataset directory ::
+
+        imagenet21k
+        ├── train
+        │   ├──class_x
+        |   |   ├── x1.jpg
+        |   |   ├── x2.jpg
+        |   |   └── ...
+        │   ├── class_y
+        |   |   ├── y1.jpg
+        |   |   ├── y2.jpg
+        |   |   └── ...
+        |   └── ...
+        └── meta
+            └── train.txt
+
 
     Args:
         data_root (str): The root directory for ``data_prefix`` and
@@ -63,12 +171,22 @@ class ImageNet21k(CustomDataset):
             Defaults to False.
         **kwargs: Other keyword arguments in :class:`CustomDataset` and
             :class:`BaseDataset`.
-    """
+
+    Examples:
+        >>> from mmpretrain.datasets import ImageNet21k
+        >>> train_dataset = ImageNet21k(data_root='data/imagenet21k', split='train')
+        >>> train_dataset
+        Dataset ImageNet21k
+            Number of samples:  14197088
+            Annotation file:    data/imagenet21k/meta/train.txt
+            Prefix of images:   data/imagenet21k/train
+    """  # noqa: E501
 
     IMG_EXTENSIONS = ('.jpg', '.jpeg', '.png', '.ppm', '.bmp', '.pgm', '.tif')
 
     def __init__(self,
                  data_root: str = '',
+                 split: str = '',
                  data_prefix: Union[str, dict] = '',
                  ann_file: str = '',
                  metainfo: Optional[dict] = None,
@@ -79,6 +197,21 @@ class ImageNet21k(CustomDataset):
                 'The `multi_label` option is not supported by now.')
         self.multi_label = multi_label
 
+        if split:
+            splits = ['train']
+            assert split in splits, \
+                f"The split must be one of {splits}, but get '{split}'.\
+                If you want to specify your own validation set or test set,\
+                please set split to None."
+
+            self.split = split
+            data_prefix = split if data_prefix == '' else data_prefix
+
+            if not ann_file:
+                _ann_path = fileio.join_path(data_root, 'meta', f'{split}.txt')
+                if fileio.exists(_ann_path):
+                    ann_file = _ann_path
+
         logger = MMLogger.get_current_instance()
 
         if not ann_file:

mmpretrain/datasets/mnist.py

@@ -7,6 +7,7 @@ import mmengine.dist as dist
 import numpy as np
 import torch
 from mmengine.fileio import LocalBackend, exists, get_file_backend, join_path
+from mmengine.logging import MMLogger
 
 from mmpretrain.registry import DATASETS
 from .base_dataset import BaseDataset
@@ -23,13 +24,11 @@ class MNIST(BaseDataset):
     https://github.com/pytorch/vision/blob/master/torchvision/datasets/mnist.py
 
     Args:
-        data_prefix (str): Prefix for data.
+        data_root (str): The root directory of the MNIST Dataset.
-        test_mode (bool): ``test_mode=True`` means in test phase.
+        split (str, optional): The dataset split, supports "train" and "test".
-            It determines to use the training set or test set.
+            Default to "train".
         metainfo (dict, optional): Meta information for dataset, such as
             categories information. Defaults to None.
-        data_root (str): The root directory for ``data_prefix``.
-            Defaults to ''.
         download (bool): Whether to download the dataset if not exists.
             Defaults to True.
         **kwargs: Other keyword arguments in :class:`BaseDataset`.
@@ -49,12 +48,29 @@ class MNIST(BaseDataset):
     METAINFO = {'classes': MNIST_CATEGORITES}
 
     def __init__(self,
-                 data_prefix: str,
-                 test_mode: bool,
-                 metainfo: Optional[dict] = None,
                  data_root: str = '',
+                 split: str = 'train',
+                 metainfo: Optional[dict] = None,
                  download: bool = True,
+                 data_prefix: str = '',
+                 test_mode: bool = False,
                  **kwargs):
+
+        splits = ['train', 'test']
+        assert split in splits, \
+            f"The split must be one of {splits}, but get '{split}'"
+        self.split = split
+
+        # To handle the BC-breaking
+        if split == 'train' and test_mode:
+            logger = MMLogger.get_current_instance()
+            logger.warning('split="train" but test_mode=True. '
+                           'The training set will be used.')
+
+        if not data_root and not data_prefix:
+            raise RuntimeError('Please set ``data_root`` to'
+                               'specify the dataset path')
+
         self.download = download
         super().__init__(
             # The MNIST dataset doesn't need specify annotation file
@@ -138,13 +154,11 @@ class FashionMNIST(MNIST):
     Dataset.
 
     Args:
-        data_prefix (str): Prefix for data.
+        data_root (str): The root directory of the MNIST Dataset.
-        test_mode (bool): ``test_mode=True`` means in test phase.
+        split (str, optional): The dataset split, supports "train" and "test".
-            It determines to use the training set or test set.
+            Default to "train".
         metainfo (dict, optional): Meta information for dataset, such as
             categories information. Defaults to None.
-        data_root (str): The root directory for ``data_prefix``.
-            Defaults to ''.
         download (bool): Whether to download the dataset if not exists.
             Defaults to True.
         **kwargs: Other keyword arguments in :class:`BaseDataset`.

mmpretrain/datasets/sun397.py

@@ -7,130 +7,12 @@ from mmpretrain.registry import DATASETS
 from .base_dataset import BaseDataset
 from .categories import SUN397_CATEGORIES
 
-# Note that some images are not a jpg file although the name ends
-# with jpg and therefore cannot be read properly. So we provide
-# a list to skip these files.
-INVALID = [
-    '/a/assembly_line/sun_ajckcfldgdrdjogj.jpg',
-    '/a/auto_factory/sun_apfsprenzdnzbhmt.jpg',
-    '/b/baggage_claim/sun_avittiqqaiibgcau.jpg',
-    '/b/batters_box/sun_alqlfpgtbgggezyr.jpg',
-    '/b/bow_window/indoor/sun_ahsholsagvlrsboa.jpg',
-    '/b/bow_window/indoor/sun_aioomcoujmmcxkkx.jpg',
-    '/b/bow_window/outdoor/sun_atgtjdpqikjmllth.jpg',
-    '/c/carrousel/sun_atsgphqympojgxnc.jpg',
-    '/c/carrousel/sun_auzitjuirwolazns.jpg',
-    '/c/church/outdoor/sun_boagasgfltequmal.jpg',
-    '/c/church/outdoor/sun_brhmnwzzbkphcvfo.jpg',
-    '/c/church/outdoor/sun_byjkqzybxpjnuofa.jpg',
-    '/c/corridor/sun_aznefxvocwpgimko.jpg',
-    '/d/dentists_office/sun_aaefsoauqlcsihou.jpg',
-    '/d/diner/indoor/sun_apswilaujhntrybg.jpg',
-    '/e/elevator/door/sun_aaudobqlphijkjdv.jpg',
-    '/f/fastfood_restaurant/sun_axeniwtesffxqedr.jpg',
-    '/f/fire_station/sun_bjyapttwilyyuxqm.jpg',
-    '/f/fountain/sun_axgmpbdyvqhtkhee.jpg',
-    '/h/hospital_room/sun_ahokhhxjiclpxqqa.jpg',
-    '/o/oast_house/sun_bqsrrygxyrutgjve.jpg',
-    '/r/restaurant_patio/sun_aurwypviprwycame.jpg',
-    '/s/ski_resort/sun_bplmntyzoiobcqhp.jpg',
-    '/w/wine_cellar/bottle_storage/sun_afmzwxkzmxkbamqi.jpg',
-    '/w/wine_cellar/bottle_storage/sun_ahyymswdjejrbhyb.jpg',
-    '/w/wine_cellar/bottle_storage/sun_avnttpxamufejbfe.jpg',
-    '/a/archive/sun_awgsrbljlsvhqjij.jpg',
-    '/a/art_school/sun_aabogqsjulyvmcse.jpg',
-    '/a/art_school/sun_apnzojafyvkariue.jpg',
-    '/b/ball_pit/sun_atjhwqngtoeuwhso.jpg',
-    '/b/bow_window/indoor/sun_asxvsqbexmmtqmht.jpg',
-    '/b/bow_window/indoor/sun_abeugxecxrwzmffp.jpg',
-    '/b/bow_window/outdoor/sun_auwcqhrtzkgihvlv.jpg',
-    '/b/bow_window/outdoor/sun_apnvdyecnjjmcuhi.jpg',
-    '/c/childs_room/sun_alggivksjwwiklmt.jpg',
-    '/c/control_tower/outdoor/sun_avbcxakrvpomqdgr.jpg',
-    '/d/diner/indoor/sun_ajmzozstvsxisvgx.jpg',
-    '/e/elevator/door/sun_aaqsyluqbluugqgy.jpg',
-    '/f/fastfood_restaurant/sun_aevchxlxoruhxgrb.jpg',
-    '/f/firing_range/indoor/sun_affrzvahwjorpalo.jpg',
-    '/f/formal_garden/sun_bjvrlaeatjufekft.jpg',
-    '/g/garage/indoor/sun_akbocuwclkxqlofx.jpg',
-    '/g/greenhouse/indoor/sun_addirvgtxfbndlwf.jpg',
-    '/k/kindergarden_classroom/sun_ajtpaahilrqzarri.jpg',
-    '/l/laundromat/sun_afrrjykuhhlwiwun.jpg',
-    '/m/music_studio/sun_bsntklkmwqgnjrjj.jpg',
-    '/t/track/outdoor/sun_aophkoiosslinihb.jpg',
-    '/a/archive/sun_aegmzltkiwyevpwa.jpg',
-    '/a/auto_factory/sun_aybymzvbxgvcrwgn.jpg',
-    '/b/baggage_claim/sun_atpmiqmnxjpgqsxi.jpg',
-    '/b/baggage_claim/sun_ajffcdpsvgqfzoxx.jpg',
-    '/b/bamboo_forest/sun_ausmxphosyahoyjo.jpg',
-    '/b/batters_box/sun_aaeheulsicxtxnbu.jpg',
-    '/c/carrousel/sun_arjrjcxemhttubqz.jpg',
-    '/c/chicken_coop/outdoor/sun_abcegmmdbizqkpgh.jpg',
-    '/c/control_tower/outdoor/sun_axhjfpkxdvqdfkyr.jpg',
-    '/d/diner/indoor/sun_apaotiublwqeowck.jpg',
-    '/f/fastfood_restaurant/sun_anexashcgmxdbmxq.jpg',
-    '/l/landing_deck/sun_aizahnjfkuurjibw.jpg',
-    '/n/nuclear_power_plant/outdoor/sun_aoblfvgyleweqanr.jpg',
-    '/w/waiting_room/sun_aicytusmthfvqcwc.jpg',
-    '/b/bow_window/indoor/sun_asmvdfnjlulewkpr.jpg',
-    '/b/bus_interior/sun_adhktvidwzmodeou.jpg',
-    '/c/catacomb/sun_algnawesgjzzmcqd.jpg',
-    '/c/church/outdoor/sun_baihxlseimcsdhdx.jpg',
-    '/d/diner/indoor/sun_agoyalzcawgxodbm.jpg',
-    '/e/elevator_shaft/sun_awaitimkinrjaybl.jpg',
-    '/f/fastfood_restaurant/sun_aplvzfbmtqtbsvbx.jpg',
-    '/g/greenhouse/indoor/sun_bkccvyfpwetwjuhk.jpg',
-    '/c/car_interior/backseat/sun_adexwfoqdyhowxpu.jpg',
-    '/c/church/outdoor/sun_blmmweiumednscuf.jpg',
-    '/f/fire_station/sun_bibntbsuunbsdrum.jpg',
-    '/g/game_room/sun_aopfaqlllpvzhrak.jpg',
-    '/u/underwater/coral_reef/sun_biiueajvszaxqopo.jpg',
-    '/a/airplane_cabin/sun_arqyikigkyfpegug.jpg',
-    '/b/badminton_court/indoor/sun_amppvxecgtjpfold.jpg',
-    '/c/carrousel/sun_anxtrtieimkpmhvk.jpg',
-    '/c/computer_room/sun_aebgvpgtwoqbfyvl.jpg',
-    '/f/fire_escape/sun_atbraxuwwlvdoolv.jpg',
-    '/k/kasbah/sun_abxkkoielpavsouu.jpg',
-    '/t/tower/sun_bccqnzcvqkiwicjt.jpg',
-    '/a/archive/sun_afngadshxudodkct.jpg',
-    '/b/bow_window/indoor/sun_awnrlipyxpgxxgxz.jpg',
-    '/c/control_tower/outdoor/sun_arohngcbtsvbthho.jpg',
-    '/f/fire_station/sun_brbskkfgghbfvgkk.jpg',
-    '/r/restaurant_patio/sun_amjfbqzfgxarrpec.jpg',
-    '/v/vineyard/sun_bdxhnbgbnolddswz.jpg',
-    '/b/baggage_claim/sun_axrtsmillrglugia.jpg',
-    '/d/diner/indoor/sun_alaqevbwpjaqqdqz.jpg',
-    '/l/landing_deck/sun_acodgoamhgnnbmvr.jpg',
-    '/c/carrousel/sun_adsafgyrinnekycc.jpg',
-    '/c/church/outdoor/sun_bzqhuwshtdgakkay.jpg',
-    '/c/closet/sun_absahzamlrylkxyn.jpg',
-    '/f/fire_escape/sun_acdthenaosuqcoqn.jpg',
-    '/b/butchers_shop/sun_asrdgbefoszenfex.jpg',
-    '/c/church/outdoor/sun_bzfyucfrdigaqneg.jpg',
-    '/c/church/outdoor/sun_byzxhknqrejdajxi.jpg',
-    '/c/cockpit/sun_ajkulpqauavrmxae.jpg',
-    '/l/living_room/sun_aefoqbeatyufobtx.jpg',
-    '/s/supermarket/sun_attvxbzocurnddbz.jpg',
-    '/c/closet/sun_aqnutmwfkypmrnfy.jpg',
-    '/f/fire_station/sun_bttrtzktpbymxkmf.jpg',
-    '/s/shopping_mall/indoor/sun_avwzjsijaxnwuzjx.jpg',
-    '/w/windmill/sun_blvczkyqbmabzeej.jpg',
-    '/c/chicken_coop/outdoor/sun_amaonsnnkskxwmrj.jpg',
-    '/s/swimming_pool/outdoor/sun_bslaihiqlhfewtzn.jpg',
-    '/u/underwater/coral_reef/sun_bhcrnmvbgnkvcvkr.jpg',
-    '/d/dining_room/sun_azlxdhiajwrhaivq.jpg',
-    '/c/church/outdoor/sun_bnunxbznqnvgeykx.jpg',
-    '/c/corridor/sun_aspwpqqlcwzfanvl.jpg',
-    '/r/restaurant_patio/sun_awcbpizjbudjvrhs.jpg',
-    '/b/ball_pit/sun_avdnmemjrgrbkwjm.jpg',
-]
-
 
 @DATASETS.register_module()
 class SUN397(BaseDataset):
     """The SUN397 Dataset.
 
-    Support the `SUN397 Dataset <https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/>`_ Dataset.
+    Support the `SUN397 Dataset <https://vision.princeton.edu/projects/2010/SUN/>`_ Dataset.
     After downloading and decompression, the dataset directory structure is as follows.
 
     SUN397 dataset directory: ::
@@ -166,15 +48,26 @@ class SUN397(BaseDataset):
         >>> train_dataset = SUN397(data_root='data/SUN397', split='train')
         >>> train_dataset
         Dataset SUN397
-            Number of samples:  19824
+            Number of samples:  19850
             Number of categories:       397
             Root of dataset:    data/SUN397
         >>> test_dataset = SUN397(data_root='data/SUN397', split='test')
         >>> test_dataset
         Dataset SUN397
-            Number of samples:  19829
+            Number of samples:  19850
             Number of categories:       397
             Root of dataset:    data/SUN397
+
+    **Note that some images are not a jpg file although the name ends with ".jpg".
+    The backend of SUN397 should be "pillow" as below to read these images properly,**
+
+    .. code-block:: python
+
+        pipeline = [
+            dict(type='LoadImageFromFile', imdecode_backend='pillow'),
+            dict(type='RandomResizedCrop', scale=224),
+            dict(type='PackInputs')
+            ]
     """  # noqa: E501
 
     METAINFO = {'classes': SUN397_CATEGORIES}
@@ -206,8 +99,6 @@ class SUN397(BaseDataset):
         pairs = list_from_file(self.ann_file)
         data_list = []
         for pair in pairs:
-            if pair in INVALID:
-                continue
             img_path = self.backend.join_path(self.img_prefix, pair[1:])
             items = pair.split('/')
             class_name = '_'.join(items[2:-1])
@@ -217,6 +108,15 @@ class SUN397(BaseDataset):
 
         return data_list
 
+    def __getitem__(self, idx: int) -> dict:
+        try:
+            return super().__getitem__(idx)
+        except AttributeError:
+            raise RuntimeError(
+                'Some images in the SUN397 dataset are not a jpg file '
+                'although the name ends with ".jpg". The backend of SUN397 '
+                'should be "pillow" to read these images properly.')
+
     def extra_repr(self) -> List[str]:
         """The extra repr information of the dataset."""
         body = [

mmpretrain/datasets/voc.py

@@ -3,6 +3,7 @@ import xml.etree.ElementTree as ET
 from typing import List, Optional, Union
 
 from mmengine import get_file_backend, list_from_file
+from mmengine.logging import MMLogger
 
 from mmpretrain.registry import DATASETS
 from .base_dataset import expanduser
@@ -18,16 +19,22 @@ class VOC(MultiLabelDataset):
 
     VOC dataset directory: ::
 
-        VOC2007 (data_root)/
+        VOC2007
-        ├── JPEGImages (data_prefix['img_path'])
+        ├── JPEGImages
         │   ├── xxx.jpg
         │   ├── xxy.jpg
         │   └── ...
-        ├── Annotations (data_prefix['ann_path'])
+        ├── Annotations
         │   ├── xxx.xml
         │   ├── xxy.xml
         │   └── ...
-        └── ImageSets (directory contains various imageset file)
+        └── ImageSets
+            └── Main
+                ├── train.txt
+                ├── val.txt
+                ├── trainval.txt
+                ├── test.txt
+                └── ...
 
     Extra difficult label is in VOC annotations, we will use
     `gt_label_difficult` to record the difficult labels in each sample
@@ -37,42 +44,85 @@ class VOC(MultiLabelDataset):
 
     Args:
         data_root (str): The root directory for VOC dataset.
-        image_set_path (str): The path of image set, The file which
+        split (str, optional): The dataset split, supports "train",
+            "val", "trainval", and "test". Default to "trainval".
+        image_set_path (str, optional): The path of image set, The file which
             lists image ids of the sub dataset, and this path is relative
-            to ``data_root``.
+            to ``data_root``. Default to ''.
         data_prefix (dict): Prefix for data and annotation, keyword
             'img_path' and 'ann_path' can be set. Defaults to be
             ``dict(img_path='JPEGImages', ann_path='Annotations')``.
-        test_mode (bool): ``test_mode=True`` means in test phase.
-            It determines to use the training set or test set.
         metainfo (dict, optional): Meta information for dataset, such as
             categories information. Defaults to None.
         **kwargs: Other keyword arguments in :class:`BaseDataset`.
+
+    Examples:
+        >>> from mmpretrain.datasets import VOC
+        >>> train_dataset = VOC(data_root='data/VOC2007', split='trainval')
+        >>> train_dataset
+        Dataset VOC
+            Number of samples:  5011
+            Number of categories:       20
+            Prefix of dataset:  data/VOC2007
+            Path of image set:  data/VOC2007/ImageSets/Main/trainval.txt
+            Prefix of images:   data/VOC2007/JPEGImages
+            Prefix of annotations:      data/VOC2007/Annotations
+        >>> test_dataset = VOC(data_root='data/VOC2007', split='test')
+        >>> test_dataset
+        Dataset VOC
+            Number of samples:  4952
+            Number of categories:       20
+            Prefix of dataset:  data/VOC2007
+            Path of image set:  data/VOC2007/ImageSets/Main/test.txt
+            Prefix of images:   data/VOC2007/JPEGImages
+            Prefix of annotations:      data/VOC2007/Annotations
     """  # noqa: E501
 
     METAINFO = {'classes': VOC2007_CATEGORIES}
 
     def __init__(self,
                  data_root: str,
-                 image_set_path: str,
+                 split: str = 'trainval',
+                 image_set_path: str = '',
                  data_prefix: Union[str, dict] = dict(
                      img_path='JPEGImages', ann_path='Annotations'),
                  test_mode: bool = False,
                  metainfo: Optional[dict] = None,
                  **kwargs):
+
+        self.backend = get_file_backend(data_root, enable_singleton=True)
+
+        if split:
+            splits = ['train', 'val', 'trainval', 'test']
+            assert split in splits, \
+                f"The split must be one of {splits}, but get '{split}'"
+            self.split = split
+
+            if not data_prefix:
+                data_prefix = dict(
+                    img_path='JPEGImages', ann_path='Annotations')
+            if not image_set_path:
+                image_set_path = self.backend.join_path(
+                    'ImageSets', 'Main', f'{split}.txt')
+
+        # To handle the BC-breaking
+        if (split == 'train' or split == 'trainval') and test_mode:
+            logger = MMLogger.get_current_instance()
+            logger.warning(f'split="{split}" but test_mode=True. '
+                           f'The {split} set will be used.')
+
         if isinstance(data_prefix, str):
             data_prefix = dict(img_path=expanduser(data_prefix))
         assert isinstance(data_prefix, dict) and 'img_path' in data_prefix, \
             '`data_prefix` must be a dict with key img_path'
 
-        if test_mode is False:
+        if (split and split not in ['val', 'test']) or not test_mode:
             assert 'ann_path' in data_prefix and data_prefix[
                 'ann_path'] is not None, \
-                '"ann_path" must be set in `data_prefix` if `test_mode` is' \
+                '"ann_path" must be set in `data_prefix`' \
-                ' False.'
+                'when validation or test set is used.'
 
         self.data_root = data_root
-        self.backend = get_file_backend(data_root, enable_singleton=True)
         self.image_set_path = self.backend.join_path(data_root, image_set_path)
 
         super().__init__(

tests/test_datasets/test_datasets.py

@@ -290,41 +290,169 @@ class TestCustomDataset(TestBaseDataset):
 class TestImageNet(TestCustomDataset):
     DATASET_TYPE = 'ImageNet'
 
-    DEFAULT_ARGS = dict(data_root=ASSETS_ROOT, ann_file='ann.txt')
+    @classmethod
+    def setUpClass(cls) -> None:
+        super().setUpClass()
+
+        tmpdir = tempfile.TemporaryDirectory()
+        cls.tmpdir = tmpdir
+        cls.root = tmpdir.name
+        cls.meta_folder = 'meta'
+        cls.train_file = 'train.txt'
+        cls.val_file = 'val.txt'
+        cls.test_file = 'test.txt'
+        cls.categories = ['cat', 'dog']
+
+        os.mkdir(osp.join(cls.root, cls.meta_folder))
+
+        cls.DEFAULT_ARGS = dict(data_root=cls.root, split='train')
+
+        with open(osp.join(cls.root, cls.meta_folder, cls.train_file),
+                  'w') as f:
+            f.write('\n'.join([
+                '1.jpg 0',
+                '2.jpg 1',
+                '3.jpg 1',
+            ]))
+
+        with open(osp.join(cls.root, cls.meta_folder, cls.val_file), 'w') as f:
+            f.write('\n'.join([
+                '11.jpg 0',
+                '22.jpg 1',
+            ]))
+
+        with open(osp.join(cls.root, cls.meta_folder, cls.test_file),
+                  'w') as f:
+            f.write('\n'.join([
+                'aa.jpg',
+                'bb.jpg',
+            ]))
+
+    def test_initialize(self):
+        super().test_initialize()
+
+        dataset_class = DATASETS.get(self.DATASET_TYPE)
+
+        # Test invalid split
+        with self.assertRaisesRegex(AssertionError, 'The split must be'):
+            cfg = {**self.DEFAULT_ARGS}
+            cfg['split'] = 'unknown'
+            dataset_class(**cfg)
+
+        # Test valid splits
+        splits = ['train', 'val']
+        for split in splits:
+            cfg = {**self.DEFAULT_ARGS}
+            cfg['split'] = split
+            cfg['classes'] = self.categories
+            dataset = dataset_class(**cfg)
+            self.assertEqual(dataset.data_root, self.root)
+
+        # Test split="test"
+        cfg = {**self.DEFAULT_ARGS}
+        cfg['split'] = 'test'
+        logger = MMLogger.get_current_instance()
+        with self.assertLogs(logger, 'INFO') as log:
+            dataset = dataset_class(**cfg)
+            self.assertFalse(dataset.with_label)
+        self.assertIn('Since the ImageNet1k test set', log.output[0])
 
     def test_load_data_list(self):
         dataset_class = DATASETS.get(self.DATASET_TYPE)
 
-        # test classes number
+        # Test default behavior
-        cfg = {
+        dataset = dataset_class(**self.DEFAULT_ARGS)
-            **self.DEFAULT_ARGS,
+        self.assertEqual(len(dataset), 3)
-            'data_prefix': ASSETS_ROOT,
+
-            'ann_file': '',
+        data_info = dataset[0]
-        }
+        self.assertEqual(data_info['img_path'],
-        with self.assertRaisesRegex(
+                         osp.join(self.root, 'train', '1.jpg'))
-                AssertionError, r"\(2\) doesn't match .* classes \(1000\)"):
+        self.assertEqual(data_info['gt_label'], 0)
-            dataset_class(**cfg)
+
+        # Test split="val"
+        cfg = {**self.DEFAULT_ARGS, 'split': 'val'}
+        dataset = dataset_class(**cfg)
+        self.assertEqual(len(dataset), 2)
+
+        data_info = dataset[0]
+        self.assertEqual(data_info['img_path'],
+                         osp.join(self.root, 'val', '11.jpg'))
+        self.assertEqual(data_info['gt_label'], 0)
+
+        # Test split="test"
+        cfg = {**self.DEFAULT_ARGS, 'split': 'test'}
+        dataset = dataset_class(**cfg)
+        self.assertEqual(len(dataset), 2)
+
+        data_info = dataset[0]
+        self.assertEqual(data_info['img_path'],
+                         osp.join(self.root, 'test', 'aa.jpg'))
 
         # test override classes
         cfg = {
             **self.DEFAULT_ARGS,
-            'data_prefix': ASSETS_ROOT,
             'classes': ['cat', 'dog'],
-            'ann_file': '',
         }
         dataset = dataset_class(**cfg)
         self.assertEqual(len(dataset), 3)
         self.assertEqual(dataset.CLASSES, ('cat', 'dog'))
 
+    def test_extra_repr(self):
+        dataset_class = DATASETS.get(self.DATASET_TYPE)
+        cfg = {**self.DEFAULT_ARGS}
+        dataset = dataset_class(**cfg)
+
+        self.assertIn(f'Root of dataset: \t{dataset.data_root}', repr(dataset))
+
 
 class TestImageNet21k(TestCustomDataset):
     DATASET_TYPE = 'ImageNet21k'
 
-    DEFAULT_ARGS = dict(
+    @classmethod
-        data_root=ASSETS_ROOT, classes=['cat', 'dog'], ann_file='ann.txt')
+    def setUpClass(cls) -> None:
+        super().setUpClass()
+
+        tmpdir = tempfile.TemporaryDirectory()
+        cls.tmpdir = tmpdir
+        cls.root = tmpdir.name
+        cls.meta_folder = 'meta'
+        cls.train_file = 'train.txt'
+
+        os.mkdir(osp.join(cls.root, cls.meta_folder))
+
+        with open(osp.join(cls.root, cls.meta_folder, cls.train_file),
+                  'w') as f:
+            f.write('\n'.join([
+                'cat/a.jpg 0',
+                'cat/b.jpg 0',
+                'dog/a.jpg 1',
+                'dog/b.jpg 1',
+            ]))
+
+        cls.DEFAULT_ARGS = dict(
+            data_root=cls.root,
+            classes=['cat', 'dog'],
+            ann_file='meta/train.txt')
+
+    def test_initialize(self):
+        super().test_initialize()
+
+        dataset_class = DATASETS.get(self.DATASET_TYPE)
+
+        # Test invalid split
+        with self.assertRaisesRegex(AssertionError, 'The split must be'):
+            cfg = {**self.DEFAULT_ARGS}
+            cfg['split'] = 'unknown'
+            dataset_class(**cfg)
+
+        # Test valid splits
+        cfg = {**self.DEFAULT_ARGS}
+        cfg['split'] = 'train'
+        dataset = dataset_class(**cfg)
+        self.assertEqual(dataset.split, 'train')
+        self.assertEqual(dataset.data_root, self.root)
 
     def test_load_data_list(self):
-        super().test_initialize()
         dataset_class = DATASETS.get(self.DATASET_TYPE)
 
         # The multi_label option is not implemented not.
@@ -333,11 +461,14 @@ class TestImageNet21k(TestCustomDataset):
             dataset_class(**cfg)
 
         # Warn about ann_file
-        cfg = {**self.DEFAULT_ARGS, 'ann_file': ''}
+        cfg = {**self.DEFAULT_ARGS, 'ann_file': '', 'lazy_init': True}
+        ann_path = osp.join(self.root, self.meta_folder, self.train_file)
+        os.rename(ann_path, ann_path + 'copy')
         logger = MMLogger.get_current_instance()
-        with self.assertLogs(logger, 'WARN') as log:
+        with self.assertLogs(logger, 'INFO') as log:
             dataset_class(**cfg)
         self.assertIn('specify the `ann_file`', log.output[0])
+        os.rename(ann_path + 'copy', ann_path)
 
         # Warn about classes
         cfg = {**self.DEFAULT_ARGS, 'classes': None}
@@ -345,6 +476,11 @@ class TestImageNet21k(TestCustomDataset):
             dataset_class(**cfg)
         self.assertIn('specify the `classes`', log.output[0])
 
+        # Test split='train'
+        cfg = {**self.DEFAULT_ARGS, 'split': 'train', 'classes': None}
+        dataset = dataset_class(**self.DEFAULT_ARGS)
+        self.assertEqual(len(dataset), 4)
+
 
 class TestPlaces205(TestCustomDataset):
     DATASET_TYPE = 'Places205'
@@ -599,17 +735,12 @@ class TestVOC(TestBaseDataset):
         cls.tmpdir = tmpdir
         data_root = tmpdir.name
 
-        cls.DEFAULT_ARGS = dict(
+        cls.DEFAULT_ARGS = dict(data_root=data_root, split='trainval')
-            data_root=data_root,
-            image_set_path='ImageSets/train.txt',
-            data_prefix=dict(img_path='JPEGImages', ann_path='Annotations'),
-            pipeline=[],
-            test_mode=False)
 
         cls.image_folder = osp.join(data_root, 'JPEGImages')
         cls.ann_folder = osp.join(data_root, 'Annotations')
-        cls.image_set_folder = osp.join(data_root, 'ImageSets')
+        cls.image_set_folder = osp.join(data_root, 'ImageSets', 'Main')
-        os.mkdir(cls.image_set_folder)
+        os.makedirs(cls.image_set_folder)
         os.mkdir(cls.image_folder)
         os.mkdir(cls.ann_folder)
 
@@ -621,7 +752,7 @@ class TestVOC(TestBaseDataset):
         train_list = [i for i in range(0, 4)]
         test_list = [i for i in range(4, 6)]
 
-        with open(osp.join(cls.image_set_folder, 'train.txt'), 'w') as f:
+        with open(osp.join(cls.image_set_folder, 'trainval.txt'), 'w') as f:
             for train_item in train_list:
                 f.write(str(train_item) + '\n')
         with open(osp.join(cls.image_set_folder, 'test.txt'), 'w') as f:
@@ -677,6 +808,31 @@ class TestVOC(TestBaseDataset):
         with self.assertRaisesRegex(ValueError, "type <class 'dict'>"):
             dataset_class(**cfg)
 
+        # Test invalid split
+        with self.assertRaisesRegex(AssertionError, 'The split must be'):
+            cfg = {**self.DEFAULT_ARGS}
+            cfg['split'] = 'unknown'
+            dataset_class(**cfg)
+
+        # Test valid splits
+        splits = ['trainval', 'test']
+        for split in splits:
+            cfg = {**self.DEFAULT_ARGS}
+            cfg['split'] = split
+            dataset = dataset_class(**cfg)
+            self.assertEqual(dataset.split, split)
+
+        # Test split='trainval' and test_mode = True
+        logger = MMLogger.get_current_instance()
+        with self.assertLogs(logger, 'WARN') as log:
+            cfg = {**self.DEFAULT_ARGS}
+            cfg['split'] = 'trainval'
+            cfg['test_mode'] = True
+            dataset = dataset_class(**cfg)
+            self.assertEqual(dataset.split, 'trainval')
+            self.assertEqual(dataset.test_mode, True)
+        self.assertIn('The trainval set will be used', log.output[0])
+
     def test_get_cat_ids(self):
         dataset_class = DATASETS.get(self.DATASET_TYPE)
         cfg = {'classes': self.fake_classes, **self.DEFAULT_ARGS}
@@ -710,9 +866,8 @@ class TestVOC(TestBaseDataset):
         self.assertEqual(data_info['img_path'], f'{fake_img_path}.jpg')
         self.assertEqual(set(data_info['gt_label']), set(self.fake_labels[0]))
 
-        # Test with test_mode=True
+        # Test with split='test'
-        cfg['image_set_path'] = 'ImageSets/test.txt'
+        cfg['split'] = 'test'
-        cfg['test_mode'] = True
         dataset = dataset_class(**cfg)
         self.assertEqual(len(dataset), 2)
 
@@ -722,7 +877,8 @@ class TestVOC(TestBaseDataset):
         self.assertEqual(set(data_info['gt_label']), set(self.fake_labels[4]))
 
         # Test with test_mode=True and ann_path = None
-        cfg['image_set_path'] = 'ImageSets/test.txt'
+        cfg['split'] = ''
+        cfg['image_set_path'] = 'ImageSets/Main/test.txt'
         cfg['test_mode'] = True
         cfg['data_prefix'] = 'JPEGImages'
         dataset = dataset_class(**cfg)
@@ -771,9 +927,9 @@ class TestMNIST(TestBaseDataset):
 
         tmpdir = tempfile.TemporaryDirectory()
         cls.tmpdir = tmpdir
+        cls.root = tmpdir.name
         data_prefix = tmpdir.name
-        cls.DEFAULT_ARGS = dict(
+        cls.DEFAULT_ARGS = dict(data_root=cls.root, split='train')
-            data_prefix=data_prefix, pipeline=[], test_mode=False)
 
         dataset_class = DATASETS.get(cls.DATASET_TYPE)
 
@@ -805,6 +961,39 @@ class TestMNIST(TestBaseDataset):
             with open(file, 'wb') as f:
                 f.write(data)
 
+    def test_initialize(self):
+        dataset_class = DATASETS.get(self.DATASET_TYPE)
+
+        # Test with invalid split
+        with self.assertRaisesRegex(AssertionError, 'The split must be'):
+            cfg = {**self.DEFAULT_ARGS}
+            cfg['split'] = 'unknown'
+            dataset_class(**cfg)
+
+        # Test with valid split
+        splits = ['train', 'test']
+        test_modes = [False, True]
+
+        for split in splits:
+            for test_mode in test_modes:
+                cfg = {**self.DEFAULT_ARGS}
+                cfg['split'] = split
+                cfg['test_mode'] = test_mode
+
+                if split == 'train' and test_mode:
+                    logger = MMLogger.get_current_instance()
+                    with self.assertLogs(logger, 'WARN') as log:
+                        dataset = dataset_class(**cfg)
+                        self.assertEqual(dataset.split, split)
+                        self.assertEqual(dataset.test_mode, test_mode)
+                        self.assertEqual(dataset.data_root, self.root)
+                    self.assertIn('training set will be used', log.output[0])
+                else:
+                    dataset = dataset_class(**cfg)
+                    self.assertEqual(dataset.split, split)
+                    self.assertEqual(dataset.test_mode, test_mode)
+                    self.assertEqual(dataset.data_root, self.root)
+
     def test_load_data_list(self):
         dataset_class = DATASETS.get(self.DATASET_TYPE)
 
@@ -817,8 +1006,8 @@ class TestMNIST(TestBaseDataset):
         np.testing.assert_equal(data_info['img'], self.fake_img)
         np.testing.assert_equal(data_info['gt_label'], self.fake_label)
 
-        # Test with test_mode=True
+        # Test with split='test'
-        cfg = {**self.DEFAULT_ARGS, 'test_mode': True}
+        cfg = {**self.DEFAULT_ARGS, 'split': 'test'}
         dataset = dataset_class(**cfg)
         self.assertEqual(len(dataset), 1)
 
@@ -829,7 +1018,7 @@ class TestMNIST(TestBaseDataset):
         # Test automatically download
         with patch('mmpretrain.datasets.mnist.download_and_extract_archive'
                    ) as mock:
-            cfg = {**self.DEFAULT_ARGS, 'lazy_init': True, 'test_mode': True}
+            cfg = {**self.DEFAULT_ARGS, 'lazy_init': True, 'split': 'test'}
             dataset = dataset_class(**cfg)
             dataset.train_list = [['invalid_train_file', None]]
             dataset.test_list = [['invalid_test_file', None]]
@@ -852,7 +1041,7 @@ class TestMNIST(TestBaseDataset):
         with self.assertRaisesRegex(RuntimeError, '`download=True`'):
             cfg = {
                 **self.DEFAULT_ARGS, 'lazy_init': True,
-                'test_mode': True,
+                'split': 'test',
                 'download': False
             }
             dataset = dataset_class(**cfg)
@@ -1946,7 +2135,7 @@ class TestSUN397(TestBaseDataset):
             f.write('\n'.join([
                 '/a/abbey/sun_aqswjsnjlrfzzhiz.jpg',
                 '/a/airplane_cabin/sun_blczihbhbntqccux.jpg',
-                '/a/assembly_line/sun_ajckcfldgdrdjogj.jpg',  # invalid
+                '/a/assembly_line/sun_ajckcfldgdrdjogj.jpg',
             ]))
 
         with open(osp.join(cls.root, cls.meta_folder, cls.test_file),
@@ -1954,8 +2143,8 @@ class TestSUN397(TestBaseDataset):
             f.write('\n'.join([
                 '/a/abbey/sun_ajkqrqitspwywirx.jpg',
                 '/a/airplane_cabin/sun_aqylhacwdsqfjuuu.jpg',
-                '/a/auto_factory/sun_apfsprenzdnzbhmt.jpg',  # invalid
+                '/a/auto_factory/sun_apfsprenzdnzbhmt.jpg',
-                '/b/baggage_claim/sun_avittiqqaiibgcau.jpg',  # invalid
+                '/b/baggage_claim/sun_avittiqqaiibgcau.jpg',
             ]))
 
     def test_initialize(self):
@@ -1981,7 +2170,7 @@ class TestSUN397(TestBaseDataset):
 
         # Test default behavior
         dataset = dataset_class(**self.DEFAULT_ARGS)
-        self.assertEqual(len(dataset), 3 - 1)
+        self.assertEqual(len(dataset), 3)
         data_info = dataset[0]
         self.assertEqual(
             data_info['img_path'],
@@ -1992,13 +2181,13 @@ class TestSUN397(TestBaseDataset):
         # Test with split='test'
         cfg = {**self.DEFAULT_ARGS, 'split': 'test'}
         dataset = dataset_class(**cfg)
-        self.assertEqual(len(dataset), 4 - 2)
+        self.assertEqual(len(dataset), 4)
         data_info = dataset[-1]
         self.assertEqual(
             data_info['img_path'],
             osp.join(self.root, self.data_prefix,
-                     'a/airplane_cabin/sun_aqylhacwdsqfjuuu.jpg'))
+                     'b/baggage_claim/sun_avittiqqaiibgcau.jpg'))
-        self.assertEqual(data_info['gt_label'], 1)
+        self.assertEqual(data_info['gt_label'], 26)
 
     def test_extra_repr(self):
         dataset_class = DATASETS.get(self.DATASET_TYPE)