v0.2.3, update configs

README.md

@@ -2,7 +2,7 @@
 # OpenMixup
 
 **News**
-* OpenMixup v0.2.3 is released, which supports new self-supervised and mixup methods (e.g., [A2MIM](https://arxiv.org/abs/2205.13943)), and adds new features as [#6](https://github.com/Westlake-AI/openmixup/issues/6). The online document is available.
+* OpenMixup v0.2.3 is released, which supports new self-supervised and mixup methods (e.g., [A2MIM](https://arxiv.org/abs/2205.13943)) and backbones ([UniFormer](https://arxiv.org/abs/2201.09450)), update the [online document](https://westlake-ai.github.io/openmixup/) and config files, and adds new features as [#6](https://github.com/Westlake-AI/openmixup/issues/6).
 * OpenMixup v0.2.2 is released, which supports new self-supervised methods ([BarlowTwins](https://arxiv.org/abs/2103.03230), [SimMIM](https://arxiv.org/abs/2111.09886), etc.), backbones ([ConvMixer](https://arxiv.org/pdf/2201.09792.pdf), [MLPMixer](https://arxiv.org/pdf/2105.01601.pdf), [VAN](https://arxiv.org/pdf/2202.09741v2.pdf), etc.), and losses as [#5](https://github.com/Westlake-AI/openmixup/issues/5).
 * OpenMixup v0.2.1 is released, which supports new methods as [#4](https://github.com/Westlake-AI/openmixup/issues/4) (bugs fixed).
 * OpenMixup v0.2.0 is released, which supports new features as [#3](https://github.com/Westlake-AI/openmixup/issues/3). We have reorganized configs and fixed bugs.

configs/classification/_base_/datasets/imagenet/swin_sz224_4xbs256.py

@@ -28,9 +28,6 @@
 data_train_root = 'data/ImageNet/train'
 data_test_list = 'data/meta/ImageNet/val_labeled.txt'
 data_test_root = 'data/ImageNet/val/'
-# Notice: Though official DeiT settings use `RepeatAugment`, we achieve competitive performances
-#   without it. This repo removes `RepeatAugment`.
-sampler = "DistributedSampler"
 
 dataset_type = 'ClassificationDataset'
 img_norm_cfg = dict(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])

configs/classification/_base_/datasets/imagenet/swin_sz224_8xbs128.py

@@ -28,9 +28,6 @@
 data_train_root = 'data/ImageNet/train'
 data_test_list = 'data/meta/ImageNet/val_labeled.txt'
 data_test_root = 'data/ImageNet/val/'
-# Notice: Though official DeiT settings use `RepeatAugment`, we achieve competitive performances
-#   without it. This repo removes `RepeatAugment`.
-sampler = "DistributedSampler"
 
 dataset_type = 'ClassificationDataset'
 img_norm_cfg = dict(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])

configs/classification/_base_/datasets/imagenet/swin_sz384_8xbs64.py

@@ -0,0 +1,88 @@
+# Refers to `_RAND_INCREASING_TRANSFORMS` in pytorch-image-models
+rand_increasing_policies = [
+    dict(type='AutoContrast'),
+    dict(type='Equalize'),
+    dict(type='Invert'),
+    dict(type='Rotate', magnitude_key='angle', magnitude_range=(0, 30)),
+    dict(type='Posterize', magnitude_key='bits', magnitude_range=(4, 0)),
+    dict(type='Solarize', magnitude_key='thr', magnitude_range=(256, 0)),
+    dict(type='SolarizeAdd', magnitude_key='magnitude', magnitude_range=(0, 110)),
+    dict(type='ColorTransform', magnitude_key='magnitude', magnitude_range=(0, 0.9)),
+    dict(type='Contrast', magnitude_key='magnitude', magnitude_range=(0, 0.9)),
+    dict(type='Brightness', magnitude_key='magnitude', magnitude_range=(0, 0.9)),
+    dict(type='Sharpness', magnitude_key='magnitude', magnitude_range=(0, 0.9)),
+    dict(type='Shear',
+        magnitude_key='magnitude', magnitude_range=(0, 0.3), direction='horizontal'),
+    dict(type='Shear',
+        magnitude_key='magnitude', magnitude_range=(0, 0.3), direction='vertical'),
+    dict(type='Translate',
+        magnitude_key='magnitude', magnitude_range=(0, 0.45), direction='horizontal'),
+    dict(type='Translate',
+        magnitude_key='magnitude', magnitude_range=(0, 0.45), direction='vertical'),
+]
+
+# dataset settings
+data_source_cfg = dict(type='ImageNet')
+# ImageNet dataset
+data_train_list = 'data/meta/ImageNet/train_labeled_full.txt'
+data_train_root = 'data/ImageNet/train'
+data_test_list = 'data/meta/ImageNet/val_labeled.txt'
+data_test_root = 'data/ImageNet/val/'
+
+dataset_type = 'ClassificationDataset'
+img_norm_cfg = dict(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+train_pipeline = [
+    dict(type='RandomResizedCrop', size=384, interpolation=3),  # bicubic
+    dict(type='RandomHorizontalFlip'),
+    dict(type='RandAugment',
+        policies=rand_increasing_policies,
+        num_policies=2, total_level=10,
+        magnitude_level=9, magnitude_std=0.5,  # DeiT or Swin
+        hparams=dict(
+            pad_val=[104, 116, 124], interpolation='bicubic')),
+    dict(
+        type='RandomErasing_numpy',  # before ToTensor and Normalize
+        erase_prob=0.25,
+        mode='rand', min_area_ratio=0.02, max_area_ratio=1 / 3,
+        fill_color=[104, 116, 124], fill_std=[58, 57, 57]),  # RGB
+]
+test_pipeline = [
+    dict(type='Resize', size=384, interpolation=3),  # 1.0
+    dict(type='CenterCrop', size=384),
+    dict(type='ToTensor'),
+    dict(type='Normalize', **img_norm_cfg),
+]
+# prefetch
+prefetch = True
+if not prefetch:
+    train_pipeline.extend([dict(type='ToTensor'), dict(type='Normalize', **img_norm_cfg)])
+
+data = dict(
+    imgs_per_gpu=64,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_source=dict(
+            list_file=data_train_list, root=data_train_root,
+            **data_source_cfg),
+        pipeline=train_pipeline,
+        prefetch=prefetch,
+    ),
+    val=dict(
+        type=dataset_type,
+        data_source=dict(
+            list_file=data_test_list, root=data_test_root, **data_source_cfg),
+        pipeline=test_pipeline,
+        prefetch=False,
+    ))
+
+# validation hook
+evaluation = dict(
+    initial=False,
+    interval=1,
+    imgs_per_gpu=128,
+    workers_per_gpu=4,
+    eval_param=dict(topk=(1, 5)))
+
+# checkpoint
+checkpoint_config = dict(interval=10, max_keep_ckpts=1)

configs/classification/_base_/datasets/imagenet/van_sz224_8xbs128.py

@@ -40,7 +40,6 @@
         magnitude_level=9, magnitude_std=0.5,
         hparams=dict(
             pad_val=[104, 116, 124], interpolation='bicubic')),
-    dict(type='ColorJitter', brightness=0.4, contrast=0.4, saturation=0.4),
     dict(
         type='RandomErasing_numpy',  # before ToTensor and Normalize
         erase_prob=0.25,

configs/classification/_base_/default_runtime.py

@@ -6,7 +6,7 @@
     interval=50,
     hooks=[
         dict(type='TextLoggerHook'),
-        dict(type='TensorboardLoggerHook')  # Bug: remove TensorboardLoggerHook in PyTorch1.10
+        # dict(type='TensorboardLoggerHook')  # Bug: remove TensorboardLoggerHook in PyTorch1.10
     ])
 # yapf:enable
 # runtime settings

configs/classification/_base_/models/convmixer/convmixer_1024_d20.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='ConvMixer',
+        arch='1024/20',
+        act_cfg=dict(type='GELU'),
+    ),
+    head=dict(
+        type='ClsMixupHead',  # mixup CE + label smooth
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=True,
+        in_channels=1024, num_classes=1000)
+)

configs/classification/_base_/models/convmixer/convmixer_1536_d20.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='ConvMixer',
+        arch='1536/20',
+        act_cfg=dict(type='GELU'),
+    ),
+    head=dict(
+        type='ClsMixupHead',  # mixup CE + label smooth
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=True,
+        in_channels=1536, num_classes=1000)
+)

configs/classification/_base_/models/convmixer/convmixer_768_d32.py

@@ -0,0 +1,19 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='ConvMixer',
+        arch='768/32',
+        act_cfg=dict(type='ReLU'),
+    ),
+    head=dict(
+        type='ClsMixupHead',  # mixup CE + label smooth
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=True,
+        in_channels=768, num_classes=1000)
+)

configs/classification/_base_/models/convnext/convnext_base.py

@@ -0,0 +1,22 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='ConvNeXt',
+        arch='base',
+        out_indices=(3,),  # x-1: stage-x
+        act_cfg=dict(type='GELU'),
+        drop_path_rate=0.5,
+        gap_before_final_norm=True,
+    ),
+    head=dict(
+        type='ClsMixupHead',
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=False,
+        in_channels=1024, num_classes=1000)
+)

configs/classification/_base_/models/convnext/convnext_large.py

@@ -0,0 +1,22 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='ConvNeXt',
+        arch='large',
+        out_indices=(3,),  # x-1: stage-x
+        act_cfg=dict(type='GELU'),
+        drop_path_rate=0.5,
+        gap_before_final_norm=True,
+    ),
+    head=dict(
+        type='ClsMixupHead',
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=False,
+        in_channels=1536, num_classes=1000)
+)

configs/classification/_base_/models/convnext/convnext_small.py

@@ -0,0 +1,22 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='ConvNeXt',
+        arch='small',
+        out_indices=(3,),  # x-1: stage-x
+        act_cfg=dict(type='GELU'),
+        drop_path_rate=0.4,
+        gap_before_final_norm=True,
+    ),
+    head=dict(
+        type='ClsMixupHead',
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=False,
+        in_channels=768, num_classes=1000)
+)

configs/classification/_base_/models/convnext/convnext_tiny.py

@@ -0,0 +1,22 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='ConvNeXt',
+        arch='tiny',
+        out_indices=(3,),  # x-1: stage-x
+        act_cfg=dict(type='GELU'),
+        drop_path_rate=0.1,
+        gap_before_final_norm=True,
+    ),
+    head=dict(
+        type='ClsMixupHead',
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=False,
+        in_channels=768, num_classes=1000)
+)

configs/classification/_base_/models/convnext/convnext_xlarge.py

@@ -0,0 +1,22 @@
+# model settings
+model = dict(
+    type='MixUpClassification',
+    pretrained=None,
+    alpha=[0.8, 1.0,],
+    mix_mode=["mixup", "cutmix",],
+    mix_args=dict(),
+    backbone=dict(
+        type='ConvNeXt',
+        arch='xlarge',
+        out_indices=(3,),  # x-1: stage-x
+        act_cfg=dict(type='GELU'),
+        drop_path_rate=0.5,
+        gap_before_final_norm=True,
+    ),
+    head=dict(
+        type='ClsMixupHead',
+        loss=dict(type='LabelSmoothLoss',
+            label_smooth_val=0.1, num_classes=1000, mode='original', loss_weight=1.0),
+        with_avg_pool=False,
+        in_channels=2048, num_classes=1000)
+)

configs/classification/_base_/models/densenet/densenet121.py

@@ -0,0 +1,13 @@
+# model settings
+model = dict(
+    type='Classification',
+    pretrained=None,
+    backbone=dict(
+        type='DenseNet', arch='121',
+        out_indices=(3,),  # x-1: stage-x
+    ),
+    head=dict(
+        type='ClsHead',
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        with_avg_pool=True, in_channels=1024, num_classes=1000)
+)

configs/classification/_base_/models/densenet/densenet161.py

@@ -0,0 +1,13 @@
+# model settings
+model = dict(
+    type='Classification',
+    pretrained=None,
+    backbone=dict(
+        type='DenseNet', arch='161',
+        out_indices=(3,),  # x-1: stage-x
+    ),
+    head=dict(
+        type='ClsHead',
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        with_avg_pool=True, in_channels=2208, num_classes=1000)
+)

configs/classification/_base_/models/densenet/densenet169.py

@@ -0,0 +1,13 @@
+# model settings
+model = dict(
+    type='Classification',
+    pretrained=None,
+    backbone=dict(
+        type='DenseNet', arch='169',
+        out_indices=(3,),  # x-1: stage-x
+    ),
+    head=dict(
+        type='ClsHead',
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        with_avg_pool=True, in_channels=1664, num_classes=1000)
+)

configs/classification/_base_/models/densenet/densenet201.py

@@ -0,0 +1,13 @@
+# model settings
+model = dict(
+    type='Classification',
+    pretrained=None,
+    backbone=dict(
+        type='DenseNet', arch='201',
+        out_indices=(3,),  # x-1: stage-x
+    ),
+    head=dict(
+        type='ClsHead',
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        with_avg_pool=True, in_channels=1920, num_classes=1000)
+)

configs/classification/_base_/models/efficientnet/efficientnet_b0.py

@@ -0,0 +1,15 @@
+# model settings
+model = dict(
+    type='Classification',
+    pretrained=None,
+    backbone=dict(
+        type='EfficientNet',
+        arch='b0',
+        out_indices=(6,),  # x-1: stage-x
+        norm_cfg=dict(type='BN', eps=1e-3),
+    ),
+    head=dict(
+        type='ClsHead',
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        with_avg_pool=True, in_channels=1280, num_classes=1000)
+)

configs/classification/_base_/models/efficientnet/efficientnet_b1.py

@@ -0,0 +1,15 @@
+# model settings
+model = dict(
+    type='Classification',
+    pretrained=None,
+    backbone=dict(
+        type='EfficientNet',
+        arch='b1',
+        out_indices=(6,),  # x-1: stage-x
+        norm_cfg=dict(type='BN', eps=1e-3),
+    ),
+    head=dict(
+        type='ClsHead',
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        with_avg_pool=True, in_channels=1280, num_classes=1000)
+)