[Feature] Support New config type (open-mmlab#2463)

Spiideo · Jun 19, 2023 · a912a39 · a912a39
1 parent 6d9d05e
commit a912a39
Show file tree

Hide file tree

Showing 35 changed files with 1,133 additions and 13 deletions.
diff --git a/README.md b/README.md
@@ -97,7 +97,7 @@ https://user-images.githubusercontent.com/15977946/124654387-0fd3c500-ded1-11eb-
 
 ## What's New
 
-- We are excited to release **YOLOX-Pose**, a One-Stage multi-person pose estimation model based on YOLOX. Checkout our [project page](/projects/yolox-pose/) for more details.
+- We are excited to release **YOLOX-Pose**, a One-Stage multi-person pose estimation model based on YOLOX. Checkout our [project page](/projects/yolox_pose/) for more details.
 
 ![yolox-pose_intro](https://user-images.githubusercontent.com/26127467/226655503-3cee746e-6e42-40be-82ae-6e7cae2a4c7e.jpg)
 
@@ -108,15 +108,15 @@ https://user-images.githubusercontent.com/15977946/124654387-0fd3c500-ded1-11eb-
   - Build individual projects with full power of MMPose but not bound up with heavy frameworks
   - Checkout new projects:
     - [RTMPose](/projects/rtmpose/)
-    - [YOLOX-Pose](/projects/yolox-pose/)
+    - [YOLOX-Pose](/projects/yolox_pose/)
     - [MMPose4AIGC](/projects/mmpose4aigc/)
   - Become a contributors and make MMPose greater. Start your journey from the [example project](/projects/example_project/)
 
 <br/>
 
 - 2022-04-06: MMPose [v1.0.0](https://github.com/open-mmlab/mmpose/releases/tag/v1.0.0) is officially released, with the main updates including:
 
-  - Release of [YOLOX-Pose](/projects/yolox-pose/), a One-Stage multi-person pose estimation model based on YOLOX
+  - Release of [YOLOX-Pose](/projects/yolox_pose/), a One-Stage multi-person pose estimation model based on YOLOX
   - Development of [MMPose for AIGC](/projects/mmpose4aigc/) based on RTMPose, generating high-quality skeleton images for Pose-guided AIGC projects
   - Support for OpenPose-style skeleton visualization
   - More complete and user-friendly [documentation and tutorials](https://mmpose.readthedocs.io/en/latest/overview.html)

diff --git a/README_CN.md b/README_CN.md
@@ -106,15 +106,15 @@ https://user-images.githubusercontent.com/15977946/124654387-0fd3c500-ded1-11eb-
   - 通过独立项目的形式，利用 MMPose 的强大功能，同时不被代码框架所束缚
   - 最新添加的项目包括：
     - [RTMPose](/projects/rtmpose/)
-    - [YOLOX-Pose](/projects/yolox-pose/)
+    - [YOLOX-Pose](/projects/yolox_pose/)
     - [MMPose4AIGC](/projects/mmpose4aigc/)
   - 从简单的 [示例项目](/projects/example_project/) 开启您的 MMPose 代码贡献者之旅吧，让我们共同打造更好用的 MMPose！
 
 <br/>
 
 - 2022-04-06：MMPose [v1.0.0](https://github.com/open-mmlab/mmpose/releases/tag/v1.0.0) 正式发布了，主要更新包括:
 
-  - 发布了 [YOLOX-Pose](/projects/yolox-pose/)，一个基于 YOLOX 的 One-Stage 多人姿态估计模型
+  - 发布了 [YOLOX-Pose](/projects/yolox_pose/)，一个基于 YOLOX 的 One-Stage 多人姿态估计模型
   - 基于 RTMPose 开发的 [MMPose for AIGC](/projects/mmpose4aigc/)，生成高质量骨架图片用于 Pose-guided AIGC 项目
   - 支持 OpenPose 风格的骨架可视化
   - 更加完善、友好的 [文档和教程](https://mmpose.readthedocs.io/zh_CN/latest/overview.html)

diff --git a/demo/MMPose_Tutorial.ipynb b/demo/MMPose_Tutorial.ipynb
@@ -11,6 +11,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "8xX3YewOtqV0"
@@ -28,6 +29,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "bkw-kUD8t3t8"
@@ -611,6 +613,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "r2bf94XpyFnk"
@@ -743,7 +746,9 @@
         "    \"\"\"Visualize predicted keypoints (and heatmaps) of one image.\"\"\"\n",
         "\n",
         "    # predict bbox\n",
-        "    init_default_scope(detector.cfg.get('default_scope', 'mmdet'))\n",
+        "    scope = detector.cfg.get('default_scope', 'mmdet')\n",
+        "    if scope is not None:\n",
+        "        init_default_scope(scope)\n",
         "    detect_result = inference_detector(detector, img_path)\n",
         "    pred_instance = detect_result.pred_instances.cpu().numpy()\n",
         "    bboxes = np.concatenate(\n",
@@ -854,6 +859,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "42HG6DSNI0Ke"
@@ -1089,6 +1095,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "H-dMbjgnJzbH"
@@ -1100,6 +1107,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "jCu4npV2rl_Q"
@@ -1195,6 +1203,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "UmGitQZkUnom"
@@ -1495,6 +1504,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "UlD8iDZehE2S"
@@ -1655,6 +1665,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "ChVqB1oYncmo"
@@ -3533,6 +3544,7 @@
       ]
     },
     {
+      "attachments": {},
       "cell_type": "markdown",
       "metadata": {
         "id": "sdLwcaojhE2T"

diff --git a/mmpose/apis/inference.py b/mmpose/apis/inference.py
@@ -96,7 +96,9 @@ def init_model(config: Union[str, Path, Config],
     config.model.train_cfg = None
 
     # register all modules in mmpose into the registries
-    init_default_scope(config.get('default_scope', 'mmpose'))
+    scope = config.get('default_scope', 'mmpose')
+    if scope is not None:
+        init_default_scope(scope)
 
     model = build_pose_estimator(config.model)
     model = revert_sync_batchnorm(model)
@@ -149,7 +151,9 @@ def inference_topdown(model: nn.Module,
         ``data_sample.pred_instances.keypoints`` and
         ``data_sample.pred_instances.keypoint_scores``.
     """
-    init_default_scope(model.cfg.get('default_scope', 'mmpose'))
+    scope = model.cfg.get('default_scope', 'mmpose')
+    if scope is not None:
+        init_default_scope(scope)
     pipeline = Compose(model.cfg.test_dataloader.dataset.pipeline)
 
     if bboxes is None:

diff --git a/mmpose/apis/inferencers/base_mmpose_inferencer.py b/mmpose/apis/inferencers/base_mmpose_inferencer.py
@@ -223,7 +223,9 @@ def _init_pipeline(self, cfg: ConfigType) -> Callable:
             ``np.ndarray``. The returned pipeline will be used to process
             a single data.
         """
-        init_default_scope(cfg.get('default_scope', 'mmpose'))
+        scope = cfg.get('default_scope', 'mmpose')
+        if scope is not None:
+            init_default_scope(scope)
         return Compose(cfg.test_dataloader.dataset.pipeline)
 
     def update_model_visualizer_settings(self, **kwargs):

diff --git a/mmpose/configs/_base_/default_runtime.py b/mmpose/configs/_base_/default_runtime.py
@@ -0,0 +1,54 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.hooks import (CheckpointHook, DistSamplerSeedHook, IterTimerHook,
+                            LoggerHook, ParamSchedulerHook, SyncBuffersHook)
+from mmengine.runner import LogProcessor
+from mmengine.visualization import LocalVisBackend
+
+from mmpose.engine.hooks import PoseVisualizationHook
+from mmpose.visualization import PoseLocalVisualizer
+
+default_scope = None
+
+# hooks
+default_hooks = dict(
+    timer=dict(type=IterTimerHook),
+    logger=dict(type=LoggerHook, interval=50),
+    param_scheduler=dict(type=ParamSchedulerHook),
+    checkpoint=dict(type=CheckpointHook, interval=10),
+    sampler_seed=dict(type=DistSamplerSeedHook),
+    visualization=dict(type=PoseVisualizationHook, enable=False),
+)
+
+# custom hooks
+custom_hooks = [
+    # Synchronize model buffers such as running_mean and running_var in BN
+    # at the end of each epoch
+    dict(type=SyncBuffersHook)
+]
+
+# multi-processing backend
+env_cfg = dict(
+    cudnn_benchmark=False,
+    mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0),
+    dist_cfg=dict(backend='nccl'),
+)
+
+# visualizer
+vis_backends = [dict(type=LocalVisBackend)]
+visualizer = dict(
+    type=PoseLocalVisualizer, vis_backends=vis_backends, name='visualizer')
+
+# logger
+log_processor = dict(
+    type=LogProcessor, window_size=50, by_epoch=True, num_digits=6)
+log_level = 'INFO'
+load_from = None
+resume = False
+
+# file I/O backend
+backend_args = dict(backend='local')
+
+# training/validation/testing progress
+train_cfg = dict(by_epoch=True)
+val_cfg = dict()
+test_cfg = dict()