[CodeCamp2023-337] New Version of config Adapting ConvNeXt Algorithm (#1760)

* add configs\_base_\datasets\imagenet21k_bs128.py

* update convnext_base_32xb128_in1k_384px.py

* add  convnext-base_32xb128_in1k.py

* add convnext-base_32xb128_in21k.py

* add convnext-large_64xb64_in1k-384px.py

* add convnext-large_64xb64_in1k.py

* add convnext-large_64xb64_in21k.py

* add convnext-small_32xb128_in1k-384px.py

* add convnext-small_32xb128_in1k.py

* add convnext-tiny_32xb128_in1k-384px.py

* add convnext-tiny_32xb128_in1k.py

* add convnext-xlarge_64xb64_in1k-384px.py

* add convnext-xlarge_64xb64_in1k.py

* add convnext-xlarge_64xb64_in21k.py

* pre-commit check

mmpretrain/configs/_base_/datasets/imagenet21k_bs128.py

@@ -0,0 +1,35 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+# This is a BETA new format config file, and the usage may change recently.
+from mmengine.dataset import DefaultSampler
+
+from mmpretrain.datasets import (ImageNet21k, LoadImageFromFile, PackInputs,
+                                 RandomFlip, RandomResizedCrop)
+
+# dataset settings
+dataset_type = ImageNet21k
+data_preprocessor = dict(
+    num_classes=21842,
+    # RGB format normalization parameters
+    mean=[123.675, 116.28, 103.53],
+    std=[58.395, 57.12, 57.375],
+    # convert image from BGR to RGB
+    to_rgb=True,
+)
+
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(type=RandomResizedCrop, scale=224),
+    dict(type=RandomFlip, prob=0.5, direction='horizontal'),
+    dict(type=PackInputs),
+]
+
+train_dataloader = dict(
+    batch_size=128,
+    num_workers=5,
+    dataset=dict(
+        type=dataset_type,
+        data_root='data/imagenet21k',
+        split='train',
+        pipeline=train_pipeline),
+    sampler=dict(type=DefaultSampler, shuffle=True),
+)

mmpretrain/configs/convnext/convnext-base_32xb128_in1k.py

@@ -0,0 +1,28 @@
+# 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
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_224 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+from mmpretrain.engine import EMAHook
+
+# dataset setting
+train_dataloader.update(batch_size=128)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=None,
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=4e-5, priority='ABOVE_NORMAL')]
+
+# 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/convnext/convnext-base_32xb128_in21k.py

@@ -0,0 +1,27 @@
+# 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
+
+with read_base():
+    from .._base_.datasets.imagenet21k_bs128 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model setting
+model.update(head=dict(num_classes=21841))
+
+# dataset setting
+data_preprocessor.update(num_classes=21841)
+train_dataloader.update(batch_size=128)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=dict(max_norm=5.0),
+)
+
+# 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/convnext/convnext-large_64xb64_in1k-384px.py

@@ -0,0 +1,27 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_384 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# dataset setting
+train_dataloader.update(batch_size=64)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=dict(max_norm=5.0),
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=4e-5, priority='ABOVE_NORMAL')]
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+# base_batch_size = (64 GPUs) x (64 samples per GPU)
+auto_scale_lr.update(base_batch_size=4096)

mmpretrain/configs/convnext/convnext-large_64xb64_in1k.py

@@ -0,0 +1,27 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_384 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# dataset setting
+train_dataloader.update(batch_size=64)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=None,
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=1e-4, priority='ABOVE_NORMAL')]
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+# base_batch_size = (64 GPUs) x (64 samples per GPU)
+auto_scale_lr.update(base_batch_size=4096)

mmpretrain/configs/convnext/convnext-large_64xb64_in21k.py

@@ -0,0 +1,26 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .._base_.datasets.imagenet21k_bs128 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model setting
+model.update(head=dict(num_classes=21841))
+
+# dataset setting
+data_preprocessor.update(num_classes=21841)
+train_dataloader.update(batch_size=64)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=dict(max_norm=5.0),
+)
+
+# 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/convnext/convnext-small_32xb128_in1k-384px.py

@@ -0,0 +1,27 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_384 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# dataset setting
+train_dataloader.update(batch_size=128)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=dict(max_norm=5.0),
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=4e-5, priority='ABOVE_NORMAL')]
+
+# 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/convnext/convnext-small_32xb128_in1k.py

@@ -0,0 +1,27 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_224 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# dataset setting
+train_dataloader.update(batch_size=128)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=None,
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=1e-4, priority='ABOVE_NORMAL')]
+
+# 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/convnext/convnext-tiny_32xb128_in1k-384px.py

@@ -0,0 +1,27 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_384 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# dataset setting
+train_dataloader.update(batch_size=128)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=dict(max_norm=5.0),
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=4e-5, priority='ABOVE_NORMAL')]
+
+# 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/convnext/convnext-tiny_32xb128_in1k.py

@@ -0,0 +1,27 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_224 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# dataset setting
+train_dataloader.update(batch_size=128)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=None,
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=1e-4, priority='ABOVE_NORMAL')]
+
+# 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/convnext/convnext-xlarge_64xb64_in1k-384px.py

@@ -0,0 +1,27 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_384 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# dataset setting
+train_dataloader.update(batch_size=64)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=dict(max_norm=5.0),
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=4e-5, priority='ABOVE_NORMAL')]
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+# base_batch_size = (64 GPUs) x (64 samples per GPU)
+auto_scale_lr.update(base_batch_size=4096)

mmpretrain/configs/convnext/convnext-xlarge_64xb64_in1k.py

@@ -0,0 +1,27 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet_bs64_swin_224 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# dataset setting
+train_dataloader.update(batch_size=64)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=None,
+)
+
+# runtime setting
+custom_hooks = [dict(type=EMAHook, momentum=1e-4, priority='ABOVE_NORMAL')]
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR
+# based on the actual training batch size.
+# base_batch_size = (64 GPUs) x (64 samples per GPU)
+auto_scale_lr.update(base_batch_size=4096)

mmpretrain/configs/convnext/convnext-xlarge_64xb64_in21k.py

@@ -0,0 +1,28 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+from mmpretrain.engine import EMAHook
+
+with read_base():
+    from .._base_.datasets.imagenet21k_bs128 import *
+    from .._base_.default_runtime import *
+    from .._base_.models.convnext_base import *
+    from .._base_.schedules.imagenet_bs1024_adamw_swin import *
+
+# model setting
+model.update(head=dict(num_classes=21841))
+
+# dataset setting
+data_preprocessor.update(num_classes=21841)
+train_dataloader.update(batch_size=64)
+
+# schedule setting
+optim_wrapper.update(
+    optimizer=dict(lr=4e-3),
+    clip_grad=dict(max_norm=5.0),
+)
+
+# 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/convnext/convnext_base_32xb128_in1k_384px.py

@@ -25,4 +25,4 @@ custom_hooks = [dict(type=EMAHook, momentum=4e-5, priority='ABOVE_NORMAL')]
 # 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 = dict(base_batch_size=4096)
+auto_scale_lr.update(base_batch_size=4096)