Merge branch 'dev'

open-mmlab · Jan 31, 2022 · e0edffb · e0edffb
2 parents 25f5cc7 + fb6d6f1
commit e0edffb
Show file tree

Hide file tree

Showing 167 changed files with 7,472 additions and 514 deletions.
diff --git a/.dev_scripts/benchmark_regression/1-benchmark_valid.py b/.dev_scripts/benchmark_regression/1-benchmark_valid.py
@@ -128,18 +128,19 @@ def inference(config_file, checkpoint, classes, args):
 
     if args.flops:
         from mmcv.cnn.utils import get_model_complexity_info
-        if hasattr(model, 'extract_feat'):
-            model.forward = model.extract_feat
-            flops, params = get_model_complexity_info(
-                model,
-                input_shape=(3, ) + resolution,
-                print_per_layer_stat=False,
-                as_strings=args.flops_str)
-            result['flops'] = flops if args.flops_str else int(flops)
-            result['params'] = params if args.flops_str else int(params)
-        else:
-            result['flops'] = ''
-            result['params'] = ''
+        with torch.no_grad():
+            if hasattr(model, 'extract_feat'):
+                model.forward = model.extract_feat
+                flops, params = get_model_complexity_info(
+                    model,
+                    input_shape=(3, ) + resolution,
+                    print_per_layer_stat=False,
+                    as_strings=args.flops_str)
+                result['flops'] = flops if args.flops_str else int(flops)
+                result['params'] = params if args.flops_str else int(params)
+            else:
+                result['flops'] = ''
+                result['params'] = ''
 
     return result
 
@@ -199,6 +200,9 @@ def main(args):
     summary_data = {}
     for model_name, model_info in models.items():
 
+        if model_info.config is None:
+            continue
+
         config = Path(model_info.config)
         assert config.exists(), f'{model_name}: {config} not found.'
 

diff --git a/.dev_scripts/benchmark_regression/2-benchmark_test.py b/.dev_scripts/benchmark_regression/2-benchmark_test.py
@@ -163,6 +163,10 @@ def test(args):
 
     preview_script = ''
     for model_info in models.values():
+
+        if model_info.results is None:
+            continue
+
         script_path = create_test_job_batch(commands, model_info, args, port,
                                             script_name)
         preview_script = script_path or preview_script
@@ -288,6 +292,9 @@ def summary(args):
     summary_data = {}
     for model_name, model_info in models.items():
 
+        if model_info.results is None:
+            continue
+
         # Skip if not found result file.
         result_file = work_dir / model_name / 'result.pkl'
         if not result_file.exists():

diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -9,9 +9,8 @@ on:
       - 'README.md'
       - 'README_zh-CN.md'
       - 'model-index.yml'
-      - 'configs/**.md'
+      - 'configs/**'
       - 'docs/**'
-      - 'docs_zh-CN/**'
       - 'demo/**'
       - '.dev_scripts/**'
 
@@ -20,9 +19,8 @@ on:
       - 'README.md'
       - 'README_zh-CN.md'
       - 'model-index.yml'
-      - 'configs/**.md'
+      - 'configs/**'
       - 'docs/**'
-      - 'docs_zh-CN/**'
       - 'demo/**'
       - '.dev_scripts/**'
 

diff --git a/README.md b/README.md
@@ -59,6 +59,13 @@ The master branch works with **PyTorch 1.5+**.
 
 ## What's new
 
+v0.20.0 was released in 30/1/2022.
+
+Highlights of the new version:
+- Support **K-fold cross-validation**. The tutorial will be released later.
+- Support **HRNet**, **ConvNeXt**, **Twins** and **EfficientNet**.
+- Support model conversion from PyTorch to **Core ML** by a tool.
+
 v0.19.0 was released in 31/12/2021.
 
 Highlights of the new version:
@@ -68,12 +75,6 @@ Highlights of the new version:
 - Support **DeiT** & **Conformer** backbone and checkpoints.
 - Provide a **CAM visualization** tool based on [pytorch-grad-cam](https://github.com/jacobgil/pytorch-grad-cam), and detailed [user guide](https://mmclassification.readthedocs.io/en/latest/tools/visualization.html#class-activation-map-visualization)!
 
-v0.18.0 was released in 30/11/2021.
-
-Highlights of the new version:
-- Support **MLP-Mixer** backbone and provide pre-trained checkpoints.
-- Add a tool to **visualize the learning rate curve** of the training phase. Welcome to use with the [tutorial](https://mmclassification.readthedocs.io/en/latest/tools/visualization.html#learning-rate-schedule-visualization)!
-
 Please refer to [changelog.md](docs/en/changelog.md) for more details and other release history.
 
 ## Installation
@@ -123,9 +124,10 @@ Results and models are available in the [model zoo](https://mmclassification.rea
 - [x] [DeiT](https://github.com/open-mmlab/mmclassification/tree/master/configs/deit)
 - [x] [Conformer](https://github.com/open-mmlab/mmclassification/tree/master/configs/conformer)
 - [x] [T2T-ViT](https://github.com/open-mmlab/mmclassification/tree/master/configs/t2t_vit)
-- [ ] EfficientNet
-- [ ] Twins
-- [ ] HRNet
+- [x] [Twins](https://github.com/open-mmlab/mmclassification/tree/master/configs/twins)
+- [x] [EfficientNet](https://github.com/open-mmlab/mmclassification/tree/master/configs/efficientnet)
+- [x] [ConvNeXt](https://github.com/open-mmlab/mmclassification/tree/master/configs/convnext)
+- [x] [HRNet](https://github.com/open-mmlab/mmclassification/tree/master/configs/hrnet)
 
 </details>
 

diff --git a/README_zh-CN.md b/README_zh-CN.md
@@ -57,6 +57,13 @@ MMClassification 是一款基于 PyTorch 的开源图像分类工具箱，是 [O
 
 ## 更新日志
 
+2022/1/30 发布了 v0.20.0 版本
+
+新版本亮点：
+- 支持 **K 折交叉验证** 工具。相应文档会在后续添加。
+- 支持了 **HRNet**，**ConvNeXt**，**Twins** 以及 **EfficientNet** 四个主干网络，欢迎使用！
+- 支持了从 PyTorch 模型到 Core-ML 模型的转换工具。
+
 2021/12/31 发布了 v0.19.0 版本
 
 新版本亮点：
@@ -66,12 +73,6 @@ MMClassification 是一款基于 PyTorch 的开源图像分类工具箱，是 [O
 - 支持了 **DeiT** 和 **Conformer** 主干网络，并提供了预训练模型。
 - 提供了一个 **CAM 可视化** 工具。该工具基于 [pytorch-grad-cam](https://github.com/jacobgil/pytorch-grad-cam)，我们提供了详细的 [使用教程](https://mmclassification.readthedocs.io/en/latest/tools/visualization.html#class-activation-map-visualization)！
 
-2021/11/30 发布了 v0.18.0 版本
-
-新版本亮点：
-- 支持了 **MLP-Mixer** 主干网络，欢迎使用！
-- 添加了一个**可视化学习率曲线**的工具，可以参考[教程](https://mmclassification.readthedocs.io/zh_CN/latest/tools/visualization.html#id3)使用
-
 发布历史和更新细节请参考 [更新日志](docs/en/changelog.md)
 
 ## 安装
@@ -121,9 +122,10 @@ MMClassification 是一款基于 PyTorch 的开源图像分类工具箱，是 [O
 - [x] [DeiT](https://github.com/open-mmlab/mmclassification/tree/master/configs/deit)
 - [x] [Conformer](https://github.com/open-mmlab/mmclassification/tree/master/configs/conformer)
 - [x] [T2T-ViT](https://github.com/open-mmlab/mmclassification/tree/master/configs/t2t_vit)
-- [ ] EfficientNet
-- [ ] Twins
-- [ ] HRNet
+- [x] [Twins](https://github.com/open-mmlab/mmclassification/tree/master/configs/twins)
+- [x] [EfficientNet](https://github.com/open-mmlab/mmclassification/tree/master/configs/efficientnet)
+- [x] [ConvNeXt](https://github.com/open-mmlab/mmclassification/tree/master/configs/convnext)
+- [x] [HRNet](https://github.com/open-mmlab/mmclassification/tree/master/configs/hrnet)
 
 </details>
 

diff --git a/configs/_base_/models/convnext/convnext-base.py b/configs/_base_/models/convnext/convnext-base.py
@@ -0,0 +1,23 @@
+# Model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='ConvNeXt',
+        arch='base',
+        out_indices=(3, ),
+        drop_path_rate=0.5,
+        gap_before_final_norm=True,
+        init_cfg=[
+            dict(
+                type='TruncNormal',
+                layer=['Conv2d', 'Linear'],
+                std=.02,
+                bias=0.),
+            dict(type='Constant', layer=['LayerNorm'], val=1., bias=0.),
+        ]),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1024,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))
diff --git a/configs/_base_/models/convnext/convnext-large.py b/configs/_base_/models/convnext/convnext-large.py
@@ -0,0 +1,23 @@
+# Model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='ConvNeXt',
+        arch='large',
+        out_indices=(3, ),
+        drop_path_rate=0.5,
+        gap_before_final_norm=True,
+        init_cfg=[
+            dict(
+                type='TruncNormal',
+                layer=['Conv2d', 'Linear'],
+                std=.02,
+                bias=0.),
+            dict(type='Constant', layer=['LayerNorm'], val=1., bias=0.),
+        ]),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1536,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))
diff --git a/configs/_base_/models/convnext/convnext-small.py b/configs/_base_/models/convnext/convnext-small.py
@@ -0,0 +1,23 @@
+# Model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='ConvNeXt',
+        arch='small',
+        out_indices=(3, ),
+        drop_path_rate=0.4,
+        gap_before_final_norm=True,
+        init_cfg=[
+            dict(
+                type='TruncNormal',
+                layer=['Conv2d', 'Linear'],
+                std=.02,
+                bias=0.),
+            dict(type='Constant', layer=['LayerNorm'], val=1., bias=0.),
+        ]),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=768,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))
diff --git a/configs/_base_/models/convnext/convnext-tiny.py b/configs/_base_/models/convnext/convnext-tiny.py
@@ -0,0 +1,23 @@
+# Model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='ConvNeXt',
+        arch='tiny',
+        out_indices=(3, ),
+        drop_path_rate=0.1,
+        gap_before_final_norm=True,
+        init_cfg=[
+            dict(
+                type='TruncNormal',
+                layer=['Conv2d', 'Linear'],
+                std=.02,
+                bias=0.),
+            dict(type='Constant', layer=['LayerNorm'], val=1., bias=0.),
+        ]),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=768,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))
diff --git a/configs/_base_/models/convnext/convnext-xlarge.py b/configs/_base_/models/convnext/convnext-xlarge.py
@@ -0,0 +1,23 @@
+# Model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(
+        type='ConvNeXt',
+        arch='xlarge',
+        out_indices=(3, ),
+        drop_path_rate=0.5,
+        gap_before_final_norm=True,
+        init_cfg=[
+            dict(
+                type='TruncNormal',
+                layer=['Conv2d', 'Linear'],
+                std=.02,
+                bias=0.),
+            dict(type='Constant', layer=['LayerNorm'], val=1., bias=0.),
+        ]),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=2048,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+    ))
diff --git a/configs/_base_/models/efficientnet_b0.py b/configs/_base_/models/efficientnet_b0.py
@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='b0'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1280,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))
diff --git a/configs/_base_/models/efficientnet_b1.py b/configs/_base_/models/efficientnet_b1.py
@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='b1'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1280,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))
diff --git a/configs/_base_/models/efficientnet_b2.py b/configs/_base_/models/efficientnet_b2.py
@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='b2'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1408,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))
diff --git a/configs/_base_/models/efficientnet_b3.py b/configs/_base_/models/efficientnet_b3.py
@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='b3'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1536,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))
diff --git a/configs/_base_/models/efficientnet_b4.py b/configs/_base_/models/efficientnet_b4.py
@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='b4'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=1792,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))
diff --git a/configs/_base_/models/efficientnet_b5.py b/configs/_base_/models/efficientnet_b5.py
@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='b5'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=2048,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))
diff --git a/configs/_base_/models/efficientnet_b6.py b/configs/_base_/models/efficientnet_b6.py
@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='b6'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=2304,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))
diff --git a/configs/_base_/models/efficientnet_b7.py b/configs/_base_/models/efficientnet_b7.py
@@ -0,0 +1,12 @@
+# model settings
+model = dict(
+    type='ImageClassifier',
+    backbone=dict(type='EfficientNet', arch='b7'),
+    neck=dict(type='GlobalAveragePooling'),
+    head=dict(
+        type='LinearClsHead',
+        num_classes=1000,
+        in_channels=2560,
+        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
+        topk=(1, 5),
+    ))