add mmpretrain/ part

mmpretrain/configs/swin_transformer_v2/swinv2_large_w12_8xb128_in21k_192px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet21k_bs128 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.swin_transformer_v2_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model settings
+model.update(
+    backbone=dict(
+        img_size=192, drop_path_rate=0.5, window_size=[12, 12, 12, 6]),
+    head=dict(num_classes=21841),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))
+
+# dataset settings
+data_preprocessor = dict(num_classes=21841)
+
+_base_['train_pipeline'][1]['scale'] = 192  # RandomResizedCrop
+_base_['test_pipeline'][1]['scale'] = 219  # ResizeEdge
+_base_['test_pipeline'][2]['crop_size'] = 192  # CenterCrop

mmpretrain/configs/swin_transformer_v2/swinv2_large_w16_in21k_pre_16xb64_in1k_256px.py

+# Only for evaluation
+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.models import CrossEntropyLoss
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_256 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.swin_transformer_v2_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model settings
+model.update(
+    backbone=dict(
+        arch='large',
+        img_size=256,
+        window_size=[16, 16, 16, 8],
+        pretrained_window_sizes=[12, 12, 12, 6]),
+    head=dict(
+        in_channels=1536,
+        loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
+        topk=(1, 5)))

mmpretrain/configs/swin_transformer_v2/swinv2_large_w24_in21k_pre_16xb64_in1k_384px.py

+# Only for evaluation
+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.models import CrossEntropyLoss
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_384 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.swin_transformer_v2_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model settings
+model.update(
+    backbone=dict(
+        arch='large',
+        img_size=384,
+        window_size=[24, 24, 24, 12],
+        pretrained_window_sizes=[12, 12, 12, 6]),
+    head=dict(
+        in_channels=1536,
+        loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
+        topk=(1, 5)))

mmpretrain/configs/swin_transformer_v2/swinv2_small_w16_16xb64_in1k_256px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_256 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.swin_transformer_v2_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model settings
+model.update(
+    backbone=dict(
+        arch='small',
+        img_size=256,
+        drop_path_rate=0.3,
+        window_size=[16, 16, 16, 8]),
+    head=dict(in_channels=768),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))

mmpretrain/configs/swin_transformer_v2/swinv2_small_w8_16xb64_in1k_256px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_256 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.swin_transformer_v2_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model settings
+model.update(
+    backbone=dict(arch='small', img_size=256, drop_path_rate=0.3),
+    head=dict(in_channels=768),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))

mmpretrain/configs/swin_transformer_v2/swinv2_tiny_w16_16xb64_in1k_256px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_256 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.swin_transformer_v2_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model settings
+model.update(
+    backbone=dict(
+        arch='tiny',
+        img_size=256,
+        drop_path_rate=0.2,
+        window_size=[16, 16, 16, 8]),
+    head=dict(in_channels=768),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))

mmpretrain/configs/swin_transformer_v2/swinv2_tiny_w8_16xb64_in1k_256px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_256 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.swin_transformer_v2_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model settings
+model.update(
+    backbone=dict(arch='tiny', img_size=256, drop_path_rate=0.2),
+    head=dict(in_channels=768),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))

mmpretrain/configs/vision_transformer/vit_base_p16_32xb128_mae_in1k.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+from torch.optim import AdamW
+
+from mmpretrain.engine import EMAHook
+from mmpretrain.models import CutMix, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_224 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+model.update(
+    backbone=dict(drop_rate=0, drop_path_rate=0.1, init_cfg=None),
+    head=dict(loss=dict(mode='original')),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=.02),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.),
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))
+
+# dataset settings
+train_dataloader.update(batch_size=128)
+
+# schedule settings
+optim_wrapper.update(
+    optimizer=dict(
+        type=AdamW,
+        lr=1e-4 * 4096 / 256,
+        weight_decay=0.3,
+        eps=1e-8,
+        betas=(0.9, 0.95)),
+    paramwise_cfg=dict(
+        norm_decay_mult=0.0,
+        bias_decay_mult=0.0,
+        custom_keys={
+            '.cls_token': dict(decay_mult=0.0),
+            '.pos_embed': dict(decay_mult=0.0)
+        }))
+
+# runtime settings
+custom_hooks = [dict(type=EMAHook, momentum=1e-4)]
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+# base_batch_size = (32 GPUs) x (128 samples per GPU)
+auto_scale_lr.update(base_batch_size=4096)

mmpretrain/configs/vision_transformer/vit_base_p16_64xb64_in1k.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.models import Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_pil_resize_autoaug import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs4096_adamw import *
+
+# model setting
+model.update(
+    head=dict(hidden_dim=3072),
+    train_cfg=dict(augments=dict(type=Mixup, alpha=0.2)),
+)
+
+# schedule setting
+optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_base_p16_64xb64_in1k_384px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.datasets import (CenterCrop, LoadImageFromFile, PackInputs,
+                                 RandomFlip, RandomResizedCrop, ResizeEdge)
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_pil_resize import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs4096_adamw import *
+
+# model setting
+model.update(backbone=dict(img_size=384))
+
+# dataset setting
+data_preprocessor.update(
+    mean=[127.5, 127.5, 127.5],
+    std=[127.5, 127.5, 127.5],
+    # convert image from BGR to RGB
+    to_rgb=True,
+)
+
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=RandomResizedCrop, scale=384, backend='pillow'),
+    dict(type=RandomFlip, prob=0.5, direction='horizontal'),
+    dict(type=PackInputs),
+]
+
+test_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=ResizeEdge, scale=384, edge='short', backend='pillow'),
+    dict(type=CenterCrop, crop_size=384),
+    dict(type=PackInputs),
+]
+
+train_dataloader.update(dataset=dict(pipeline=train_pipeline))
+val_dataloader.update(dataset=dict(pipeline=test_pipeline))
+test_dataloader.update(dataset=dict(pipeline=test_pipeline))
+
+# schedule setting
+optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_base_p32_64xb64_in1k.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.models import CrossEntropyLoss, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_pil_resize_autoaug import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs4096_adamw import *
+
+# model setting
+model.update(
+    backbone=dict(patch_size=32),
+    head=dict(
+        hidden_dim=3072,
+        topk=(1, 5),
+    ),
+    train_cfg=dict(augments=dict(type=Mixup, alpha=0.2)),
+)
+
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
+# schedule setting
+optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_base_p32_64xb64_in1k_384px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.datasets import (CenterCrop, LoadImageFromFile, PackInputs,
+                                 RandomFlip, RandomResizedCrop, ResizeEdge)
+from mmpretrain.models import CrossEntropyLoss
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_pil_resize import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs4096_adamw import *
+
+# model setting
+model.update(
+    backbone=dict(img_size=384, patch_size=32), head=dict(topk=(1, 5)))
+
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
+# dataset setting
+data_preprocessor.update(
+    mean=[127.5, 127.5, 127.5],
+    std=[127.5, 127.5, 127.5],
+    # convert image from BGR to RGB
+    to_rgb=True,
+)
+
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=RandomResizedCrop, scale=384, backend='pillow'),
+    dict(type=RandomFlip, prob=0.5, direction='horizontal'),
+    dict(type=PackInputs),
+]
+
+test_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=ResizeEdge, scale=384, edge='short', backend='pillow'),
+    dict(type=CenterCrop, crop_size=384),
+    dict(type=PackInputs),
+]
+
+train_dataloader.update(dataset=dict(pipeline=train_pipeline))
+val_dataloader.update(dataset=dict(pipeline=test_pipeline))
+test_dataloader.update(dataset=dict(pipeline=test_pipeline))
+
+# schedule setting
+optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_large_p16_64xb64_in1k.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.models import CrossEntropyLoss, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_pil_resize_autoaug import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs4096_adamw import *
+
+# model setting
+model.update(
+    backbone=dict(arch='l'),
+    head=dict(
+        hidden_dim=3072,
+        in_channels=1024,
+        topk=(1, 5),
+    ),
+    train_cfg=dict(augments=dict(type=Mixup, alpha=0.2)),
+)
+
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
+# schedule setting
+optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_large_p16_64xb64_in1k_384px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.datasets import (CenterCrop, LoadImageFromFile, PackInputs,
+                                 RandomFlip, RandomResizedCrop, ResizeEdge)
+from mmpretrain.models import CrossEntropyLoss
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_pil_resize import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs4096_adamw import *
+
+# model setting
+model.update(
+    backbone=dict(arch='l', img_size=384),
+    head=dict(in_channels=1024, topk=(1, 5)))
+
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
+# dataset setting
+data_preprocessor.update(
+    mean=[127.5, 127.5, 127.5],
+    std=[127.5, 127.5, 127.5],
+    # convert image from BGR to RGB
+    to_rgb=True,
+)
+
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=RandomResizedCrop, scale=384, backend='pillow'),
+    dict(type=RandomFlip, prob=0.5, direction='horizontal'),
+    dict(type=PackInputs),
+]
+
+test_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=ResizeEdge, scale=384, edge='short', backend='pillow'),
+    dict(type=CenterCrop, crop_size=384),
+    dict(type=PackInputs),
+]
+
+train_dataloader.update(dataset=dict(pipeline=train_pipeline))
+val_dataloader.update(dataset=dict(pipeline=test_pipeline))
+test_dataloader.update(dataset=dict(pipeline=test_pipeline))
+
+# schedule setting
+optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_large_p32_64xb64_in1k.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.models import CrossEntropyLoss, Mixup
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_pil_resize_autoaug import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs4096_adamw import *
+
+# model setting
+model.update(
+    backbone=dict(arch='l', patch_size=32),
+    head=dict(
+        hidden_dim=3072,
+        in_channels=1024,
+        topk=(1, 5),
+    ),
+    train_cfg=dict(augments=dict(type=Mixup, alpha=0.2)),
+)
+
+loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
+# schedule setting
+optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_large_p32_64xb64_in1k_384px.py

+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.config import read_base
+
+from mmpretrain.datasets import (CenterCrop, LoadImageFromFile, PackInputs,
+                                 RandomFlip, RandomResizedCrop, ResizeEdge)
+from mmpretrain.models import CrossEntropyLoss
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_pil_resize import *
+    from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
+    from .._base_.schedules.imagenet_bs4096_adamw import *
+
+# model setting
+model.update(
+    backbone=dict(arch='l', img_size=384, patch_size=32),
+    head=dict(in_channels=1024, topk=(1, 5)))
+
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
+# dataset setting
+data_preprocessor.update(
+    mean=[127.5, 127.5, 127.5],
+    std=[127.5, 127.5, 127.5],
+    # convert image from BGR to RGB
+    to_rgb=True,
+)
+
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=RandomResizedCrop, scale=384, backend='pillow'),
+    dict(type=RandomFlip, prob=0.5, direction='horizontal'),
+    dict(type=PackInputs),
+]
+
+test_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=ResizeEdge, scale=384, edge='short', backend='pillow'),
+    dict(type=CenterCrop, crop_size=384),
+    dict(type=PackInputs),
+]
+
+train_dataloader.update(dataset=dict(pipeline=train_pipeline))
+val_dataloader.update(dataset=dict(pipeline=test_pipeline))
+test_dataloader.update(dataset=dict(pipeline=test_pipeline))
+
+# schedule setting
+optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/datasets/__init__.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmpretrain.utils.dependency import WITH_MULTIMODAL
+from .base_dataset import BaseDataset
+from .builder import build_dataset
+from .caltech101 import Caltech101
+from .cifar import CIFAR10, CIFAR100
+from .cub import CUB
+from .custom import CustomDataset
+from .dataset_wrappers import KFoldDataset
+from .dtd import DTD
+from .fgvcaircraft import FGVCAircraft
+from .flowers102 import Flowers102
+from .food101 import Food101
+from .imagenet import ImageNet, ImageNet21k
+from .inshop import InShop
+from .mnist import MNIST, FashionMNIST
+from .multi_label import MultiLabelDataset
+from .multi_task import MultiTaskDataset
+from .nlvr2 import NLVR2
+from .oxfordiiitpet import OxfordIIITPet
+from .places205 import Places205
+from .samplers import *  # noqa: F401,F403
+from .stanfordcars import StanfordCars
+from .sun397 import SUN397
+from .transforms import *  # noqa: F401,F403
+from .voc import VOC
+
+__all__ = [
+    'BaseDataset', 'CIFAR10', 'CIFAR100', 'CUB', 'Caltech101', 'CustomDataset',
+    'DTD', 'FGVCAircraft', 'FashionMNIST', 'Flowers102', 'Food101', 'ImageNet',
+    'ImageNet21k', 'InShop', 'KFoldDataset', 'MNIST', 'MultiLabelDataset',
+    'MultiTaskDataset', 'NLVR2', 'OxfordIIITPet', 'Places205', 'SUN397',
+    'StanfordCars', 'VOC', 'build_dataset'
+]
+
+if WITH_MULTIMODAL:
+    from .coco_caption import COCOCaption
+    from .coco_retrieval import COCORetrieval
+    from .coco_vqa import COCOVQA
+    from .flamingo import FlamingoEvalCOCOCaption, FlamingoEvalCOCOVQA
+    from .flickr30k_caption import Flickr30kCaption
+    from .flickr30k_retrieval import Flickr30kRetrieval
+    from .gqa_dataset import GQA
+    from .iconqa import IconQA
+    from .infographic_vqa import InfographicVQA
+    from .minigpt4_dataset import MiniGPT4Dataset
+    from .nocaps import NoCaps
+    from .ocr_vqa import OCRVQA
+    from .refcoco import RefCOCO
+    from .scienceqa import ScienceQA
+    from .textvqa import TextVQA
+    from .visual_genome import VisualGenomeQA
+    from .vizwiz import VizWiz
+    from .vsr import VSR
+
+    __all__.extend([
+        'COCOCaption', 'COCORetrieval', 'COCOVQA', 'FlamingoEvalCOCOCaption',
+        'FlamingoEvalCOCOVQA', 'Flickr30kCaption', 'Flickr30kRetrieval',
+        'RefCOCO', 'VisualGenomeQA', 'ScienceQA', 'NoCaps', 'GQA', 'TextVQA',
+        'VSR', 'VizWiz', 'OCRVQA', 'InfographicVQA', 'IconQA',
+        'MiniGPT4Dataset'
+    ])

mmpretrain/datasets/base_dataset.py

+# Copyright (c) OpenMMLab. All rights reserved.
+import os.path as osp
+from os import PathLike
+from typing import List, Optional, Sequence, Union
+
+import mmengine
+import numpy as np
+from mmengine.dataset import BaseDataset as _BaseDataset
+
+from mmpretrain.registry import DATASETS, TRANSFORMS
+
+
+def expanduser(path):
+    """Expand ~ and ~user constructions.
+
+    If user or $HOME is unknown, do nothing.
+    """
+    if isinstance(path, (str, PathLike)):
+        return osp.expanduser(path)
+    else:
+        return path
+
+
+@DATASETS.register_module()
+class BaseDataset(_BaseDataset):
+    """Base dataset for image classification task.
+
+    This dataset support annotation file in `OpenMMLab 2.0 style annotation
+    format`.
+
+    .. _OpenMMLab 2.0 style annotation format:
+        https://github.com/open-mmlab/mmengine/blob/main/docs/zh_cn/tutorials/basedataset.md
+
+    Comparing with the :class:`mmengine.BaseDataset`, this class implemented
+    several useful methods.
+
+    Args:
+        ann_file (str): Annotation file path.
+        metainfo (dict, optional): Meta information for dataset, such as class
+            information. Defaults to None.
+        data_root (str): The root directory for ``data_prefix`` and
+            ``ann_file``. Defaults to ''.
+        data_prefix (str | dict): Prefix for training data. Defaults to ''.
+        filter_cfg (dict, optional): Config for filter data. Defaults to None.
+        indices (int or Sequence[int], optional): Support using first few
+            data in annotation file to facilitate training/testing on a smaller
+            dataset. Defaults to None, which means using all ``data_infos``.
+        serialize_data (bool): Whether to hold memory using serialized objects,
+            when enabled, data loader workers can use shared RAM from master
+            process instead of making a copy. Defaults to True.
+        pipeline (Sequence): Processing pipeline. Defaults to an empty tuple.
+        test_mode (bool, optional): ``test_mode=True`` means in test phase,
+            an error will be raised when getting an item fails, ``test_mode=False``
+            means in training phase, another item will be returned randomly.
+            Defaults to False.
+        lazy_init (bool): Whether to load annotation during instantiation.
+            In some cases, such as visualization, only the meta information of
+            the dataset is needed, which is not necessary to load annotation
+            file. ``Basedataset`` can skip load annotations to save time by set
+            ``lazy_init=False``. Defaults to False.
+        max_refetch (int): If ``Basedataset.prepare_data`` get a None img.
+            The maximum extra number of cycles to get a valid image.
+            Defaults to 1000.
+        classes (str | Sequence[str], optional): Specify names of classes.
+
+            - If is string, it should be a file path, and the every line of
+              the file is a name of a class.
+            - If is a sequence of string, every item is a name of class.
+            - If is None, use categories information in ``metainfo`` argument,
+              annotation file or the class attribute ``METAINFO``.
+
+            Defaults to None.
+    """  # noqa: E501
+
+    def __init__(self,
+                 ann_file: str,
+                 metainfo: Optional[dict] = None,
+                 data_root: str = '',
+                 data_prefix: Union[str, dict] = '',
+                 filter_cfg: Optional[dict] = None,
+                 indices: Optional[Union[int, Sequence[int]]] = None,
+                 serialize_data: bool = True,
+                 pipeline: Sequence = (),
+                 test_mode: bool = False,
+                 lazy_init: bool = False,
+                 max_refetch: int = 1000,
+                 classes: Union[str, Sequence[str], None] = None):
+        if isinstance(data_prefix, str):
+            data_prefix = dict(img_path=expanduser(data_prefix))
+
+        ann_file = expanduser(ann_file)
+        metainfo = self._compat_classes(metainfo, classes)
+
+        transforms = []
+        for transform in pipeline:
+            if isinstance(transform, dict):
+                transforms.append(TRANSFORMS.build(transform))
+            else:
+                transforms.append(transform)
+
+        super().__init__(
+            ann_file=ann_file,
+            metainfo=metainfo,
+            data_root=data_root,
+            data_prefix=data_prefix,
+            filter_cfg=filter_cfg,
+            indices=indices,
+            serialize_data=serialize_data,
+            pipeline=transforms,
+            test_mode=test_mode,
+            lazy_init=lazy_init,
+            max_refetch=max_refetch)
+
+    @property
+    def img_prefix(self):
+        """The prefix of images."""
+        return self.data_prefix['img_path']
+
+    @property
+    def CLASSES(self):
+        """Return all categories names."""
+        return self._metainfo.get('classes', None)
+
+    @property
+    def class_to_idx(self):
+        """Map mapping class name to class index.
+
+        Returns:
+            dict: mapping from class name to class index.
+        """
+
+        return {cat: i for i, cat in enumerate(self.CLASSES)}
+
+    def get_gt_labels(self):
+        """Get all ground-truth labels (categories).
+
+        Returns:
+            np.ndarray: categories for all images.
+        """
+
+        gt_labels = np.array(
+            [self.get_data_info(i)['gt_label'] for i in range(len(self))])
+        return gt_labels
+
+    def get_cat_ids(self, idx: int) -> List[int]:
+        """Get category id by index.
+
+        Args:
+            idx (int): Index of data.
+
+        Returns:
+            cat_ids (List[int]): Image category of specified index.
+        """
+
+        return [int(self.get_data_info(idx)['gt_label'])]
+
+    def _compat_classes(self, metainfo, classes):
+        """Merge the old style ``classes`` arguments to ``metainfo``."""
+        if isinstance(classes, str):
+            # take it as a file path
+            class_names = mmengine.list_from_file(expanduser(classes))
+        elif isinstance(classes, (tuple, list)):
+            class_names = classes
+        elif classes is not None:
+            raise ValueError(f'Unsupported type {type(classes)} of classes.')
+
+        if metainfo is None:
+            metainfo = {}
+
+        if classes is not None:
+            metainfo = {'classes': tuple(class_names), **metainfo}
+
+        return metainfo
+
+    def full_init(self):
+        """Load annotation file and set ``BaseDataset._fully_initialized`` to
+        True."""
+        super().full_init()
+
+        #  To support the standard OpenMMLab 2.0 annotation format. Generate
+        #  metainfo in internal format from standard metainfo format.
+        if 'categories' in self._metainfo and 'classes' not in self._metainfo:
+            categories = sorted(
+                self._metainfo['categories'], key=lambda x: x['id'])
+            self._metainfo['classes'] = tuple(
+                [cat['category_name'] for cat in categories])
+
+    def __repr__(self):
+        """Print the basic information of the dataset.
+
+        Returns:
+            str: Formatted string.
+        """
+        head = 'Dataset ' + self.__class__.__name__
+        body = []
+        if self._fully_initialized:
+            body.append(f'Number of samples: \t{self.__len__()}')
+        else:
+            body.append("Haven't been initialized")
+
+        if self.CLASSES is not None:
+            body.append(f'Number of categories: \t{len(self.CLASSES)}')
+
+        body.extend(self.extra_repr())
+
+        if len(self.pipeline.transforms) > 0:
+            body.append('With transforms:')
+            for t in self.pipeline.transforms:
+                body.append(f'    {t}')
+
+        lines = [head] + [' ' * 4 + line for line in body]
+        return '\n'.join(lines)
+
+    def extra_repr(self) -> List[str]:
+        """The extra repr information of the dataset."""
+        body = []
+        body.append(f'Annotation file: \t{self.ann_file}')
+        body.append(f'Prefix of images: \t{self.img_prefix}')
+        return body

mmpretrain/datasets/builder.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmpretrain.registry import DATASETS
+
+
+def build_dataset(cfg):
+    """Build dataset.
+
+    Examples:
+        >>> from mmpretrain.datasets import build_dataset
+        >>> mnist_train = build_dataset(
+        ...     dict(type='MNIST', data_prefix='data/mnist/', test_mode=False))
+        >>> print(mnist_train)
+        Dataset MNIST
+            Number of samples:  60000
+            Number of categories:       10
+            Prefix of data:     data/mnist/
+        >>> mnist_test = build_dataset(
+        ...     dict(type='MNIST', data_prefix='data/mnist/', test_mode=True))
+        >>> print(mnist_test)
+        Dataset MNIST
+            Number of samples:  10000
+            Number of categories:       10
+            Prefix of data:     data/mnist/
+    """
+    return DATASETS.build(cfg)

mmpretrain/datasets/caltech101.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from typing import List
+
+from mmengine import get_file_backend, list_from_file
+
+from mmpretrain.registry import DATASETS
+from .base_dataset import BaseDataset
+from .categories import CALTECH101_CATEGORIES
+
+
+@DATASETS.register_module()
+class Caltech101(BaseDataset):
+    """The Caltech101 Dataset.
+
+    Support the `Caltech101 <https://data.caltech.edu/records/mzrjq-6wc02>`_ Dataset.
+    After downloading and decompression, the dataset directory structure is as follows.
+
+    Caltech101 dataset directory: ::
+
+        caltech-101
+        ├── 101_ObjectCategories
+        │   ├── class_x
+        │   │   ├── xx1.jpg
+        │   │   ├── xx2.jpg
+        │   │   └── ...
+        │   ├── class_y
+        │   │   ├── yy1.jpg
+        │   │   ├── yy2.jpg
+        │   │   └── ...
+        │   └── ...
+        ├── Annotations
+        │   ├── class_x
+        │   │   ├── xx1.mat
+        │   │   └── ...
+        │   └── ...
+        ├── meta
+        │   ├── train.txt
+        │   └── test.txt
+        └── ....
+
+    Please note that since there is no official splitting for training and
+    test set, you can use the train.txt and text.txt provided by us or
+    create your own annotation files. Here is the download
+    `link <https://download.openmmlab.com/mmpretrain/datasets/caltech_meta.zip>`_
+    for the annotations.
+
+    Args:
+        data_root (str): The root directory for the Caltech101 dataset.
+        split (str, optional): The dataset split, supports "train" and "test".
+            Default to "train".
+
+    Examples:
+        >>> from mmpretrain.datasets import Caltech101
+        >>> train_dataset = Caltech101(data_root='data/caltech-101', split='train')
+        >>> train_dataset
+        Dataset Caltech101
+            Number of samples:  3060
+            Number of categories:       102
+            Root of dataset:    data/caltech-101
+        >>> test_dataset = Caltech101(data_root='data/caltech-101', split='test')
+        >>> test_dataset
+        Dataset Caltech101
+            Number of samples:  6728
+            Number of categories:       102
+            Root of dataset:    data/caltech-101
+    """  # noqa: E501
+
+    METAINFO = {'classes': CALTECH101_CATEGORIES}
+
+    def __init__(self, data_root: str, split: str = 'train', **kwargs):
+
+        splits = ['train', 'test']
+        assert split in splits, \
+            f"The split must be one of {splits}, but get '{split}'"
+        self.split = split
+
+        self.backend = get_file_backend(data_root, enable_singleton=True)
+
+        if split == 'train':
+            ann_file = self.backend.join_path('meta', 'train.txt')
+        else:
+            ann_file = self.backend.join_path('meta', 'test.txt')
+
+        data_prefix = '101_ObjectCategories'
+        test_mode = split == 'test'
+
+        super(Caltech101, self).__init__(
+            ann_file=ann_file,
+            data_root=data_root,
+            data_prefix=data_prefix,
+            test_mode=test_mode,
+            **kwargs)
+
+    def load_data_list(self):
+        """Load images and ground truth labels."""
+
+        pairs = list_from_file(self.ann_file)
+        data_list = []
+
+        for pair in pairs:
+            path, gt_label = pair.split()
+            img_path = self.backend.join_path(self.img_prefix, path)
+            info = dict(img_path=img_path, gt_label=int(gt_label))
+            data_list.append(info)
+
+        return data_list
+
+    def extra_repr(self) -> List[str]:
+        """The extra repr information of the dataset."""
+        body = [
+            f'Root of dataset: \t{self.data_root}',
+        ]
+        return body