merge

Author: hhaAndroid

.circleci/test.yml

@@ -162,8 +162,8 @@ workflows:
             - lint
       - build_cpu:
           name: maximum_version_cpu
-          torch: 1.12.1
-          torchvision: 0.13.1
+          torch: 1.13.0
+          torchvision: 0.14.0
           python: 3.9.0
           requires:
             - minimum_version_cpu

.dev_scripts/test_init_backbone.py

@@ -9,7 +9,7 @@
 from mmengine.runner import CheckpointLoader
 from mmengine.utils import ProgressBar
 
-from mmdet.models import build_detector
+from mmdet.registry import MODELS
 
 
 def _get_config_directory():
@@ -93,7 +93,7 @@ def _check_backbone(config, print_cfg=True):
 
     First, using `CheckpointLoader.load_checkpoint` to load the checkpoint
         without loading models.
-    Then, using `build_detector` to build models, and using
+    Then, using `MODELS.build` to build models, and using
         `model.init_weights()` to initialize the parameters.
     Finally, assert weights and bias of each layer loaded from pretrained
         checkpoint are equal to the weights and bias of original checkpoint.
@@ -126,10 +126,7 @@ def _check_backbone(config, print_cfg=True):
         else:
             state_dict = checkpoint
 
-        model = build_detector(
-            cfg.model,
-            train_cfg=cfg.get('train_cfg'),
-            test_cfg=cfg.get('test_cfg'))
+        model = MODELS.build(cfg.model)
         model.init_weights()
 
         checkpoint_layers = state_dict.keys()

README.md

@@ -68,13 +68,32 @@ The master branch works with **PyTorch 1.6+**.
 - **State of the art**
 
   The toolbox stems from the codebase developed by the *MMDet* team, who won [COCO Detection Challenge](http://cocodataset.org/#detection-leaderboard) in 2018, and we keep pushing it forward.
+  The newly released [RTMDet](configs/rtmdet) also obtains new state-of-the-art results on real-time instance segmentation and rotated object detection tasks and the best parameter-accuracy trade-off on object detection.
 
 </details>
 
 Apart from MMDetection, we also released [MMEngine](https://github.com/open-mmlab/mmengine) for model training and [MMCV](https://github.com/open-mmlab/mmcv) for computer vision research, which are heavily depended on by this toolbox.
 
 ## What's New
 
+### Highlight
+
+We are excited to announce our latest work on real-time object recognition tasks, **RTMDet**, a family of fully convolutional single-stage detectors. RTMDet not only achieves the best parameter-accuracy trade-off on object detection from tiny to extra-large model sizes but also obtains new state-of-the-art performance on instance segmentation and rotated object detection tasks. Details can be found in the [technical report](https://arxiv.org/abs/2212.07784). Pre-trained models are [here](configs/rtmdet).
+
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rtmdet-an-empirical-study-of-designing-real/real-time-instance-segmentation-on-mscoco)](https://paperswithcode.com/sota/real-time-instance-segmentation-on-mscoco?p=rtmdet-an-empirical-study-of-designing-real)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rtmdet-an-empirical-study-of-designing-real/object-detection-in-aerial-images-on-dota-1)](https://paperswithcode.com/sota/object-detection-in-aerial-images-on-dota-1?p=rtmdet-an-empirical-study-of-designing-real)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rtmdet-an-empirical-study-of-designing-real/object-detection-in-aerial-images-on-hrsc2016)](https://paperswithcode.com/sota/object-detection-in-aerial-images-on-hrsc2016?p=rtmdet-an-empirical-study-of-designing-real)
+
+| Task                     | Dataset | AP                                   | FPS(TRT FP16 BS1 3090) |
+| ------------------------ | ------- | ------------------------------------ | ---------------------- |
+| Object Detection         | COCO    | 52.8                                 | 322                    |
+| Instance Segmentation    | COCO    | 44.6                                 | 188                    |
+| Rotated Object Detection | DOTA    | 78.9(single-scale)/81.3(multi-scale) | 121                    |
+
+<div align=center>
+<img src="https://user-images.githubusercontent.com/12907710/208044554-1e8de6b5-48d8-44e4-a7b5-75076c7ebb71.png"/>
+</div>
+
 **v3.0.0rc4** was released in 25/11/2022:
 
 - Support [CondInst](https://arxiv.org/abs/2003.05664)
@@ -187,6 +206,7 @@ Results and models are available in the [model zoo](docs/en/model_zoo.md).
             <li><a href="configs/deformable_detr">Deformable DETR (ICLR'2021)</a></li>
             <li><a href="configs/tood">TOOD (ICCV'2021)</a></li>
             <li><a href="configs/ddod">DDOD (ACM MM'2021)</a></li>
+            <li><a href="configs/RTMDet">RTMDet (ArXiv'2022)</a></li>
       </ul>
       </td>
       <td>
@@ -206,6 +226,7 @@ Results and models are available in the [model zoo](docs/en/model_zoo.md).
           <li><a href="configs/mask2former">Mask2Former (ArXiv'2021)</a></li>
           <li><a href="configs/condinst">CondInst (ECCV 2020)</a></li>
           <li><a href="projects/SparseInst">SparseInst (CVPR 2022)</a></li>
+          <li><a href="configs/RTMDet">RTMDet (ArXiv'2022)</a></li>
         </ul>
       </td>
       <td>

README_zh-CN.md

@@ -67,13 +67,32 @@ MMDetection 是一个基于 PyTorch 的目标检测开源工具箱。它是 [Ope
 - **性能高**
 
   MMDetection 这个算法库源自于 COCO 2018 目标检测竞赛的冠军团队 *MMDet* 团队开发的代码，我们在之后持续进行了改进和提升。
+  新发布的 [RTMDet](configs/rtmdet) 还在实时实例分割和旋转目标检测任务中取得了最先进的成果，同时也在目标检测模型中取得了最佳的的参数量和精度平衡。
 
 </details>
 
 除了 MMDetection 之外，我们还开源了深度学习训练库 [MMEngine](https://github.com/open-mmlab/mmengine) 和计算机视觉基础库 [MMCV](https://github.com/open-mmlab/mmcv)，它们是 MMDetection 的主要依赖。
 
 ## 最新进展
 
+### 亮点
+
+我们很高兴向大家介绍我们在实时目标识别任务方面的最新成果 RTMDet，包含了一系列的全卷积单阶段检测模型。 RTMDet 不仅在从 tiny 到 extra-large 尺寸的目标检测模型上上实现了最佳的参数量和精度的平衡，而且在实时实例分割和旋转目标检测任务上取得了最先进的成果。 更多细节请参阅[技术报告](https://arxiv.org/abs/2212.07784)。 预训练模型可以在[这里](configs/rtmdet)找到。
+
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rtmdet-an-empirical-study-of-designing-real/real-time-instance-segmentation-on-mscoco)](https://paperswithcode.com/sota/real-time-instance-segmentation-on-mscoco?p=rtmdet-an-empirical-study-of-designing-real)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rtmdet-an-empirical-study-of-designing-real/object-detection-in-aerial-images-on-dota-1)](https://paperswithcode.com/sota/object-detection-in-aerial-images-on-dota-1?p=rtmdet-an-empirical-study-of-designing-real)
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rtmdet-an-empirical-study-of-designing-real/object-detection-in-aerial-images-on-hrsc2016)](https://paperswithcode.com/sota/object-detection-in-aerial-images-on-hrsc2016?p=rtmdet-an-empirical-study-of-designing-real)
+
+| Task                     | Dataset | AP                                   | FPS(TRT FP16 BS1 3090) |
+| ------------------------ | ------- | ------------------------------------ | ---------------------- |
+| Object Detection         | COCO    | 52.8                                 | 322                    |
+| Instance Segmentation    | COCO    | 44.6                                 | 188                    |
+| Rotated Object Detection | DOTA    | 78.9(single-scale)/81.3(multi-scale) | 121                    |
+
+<div align=center>
+<img src="https://user-images.githubusercontent.com/12907710/208044554-1e8de6b5-48d8-44e4-a7b5-75076c7ebb71.png"/>
+</div>
+
 **v3.0.0rc4** 版本已经在 2022.11.25 发布：
 
 - 支持了 [CondInst](https://arxiv.org/abs/2003.05664)
@@ -188,6 +207,7 @@ MMDetection 是一个基于 PyTorch 的目标检测开源工具箱。它是 [Ope
             <li><a href="configs/deformable_detr">Deformable DETR (ICLR'2021)</a></li>
             <li><a href="configs/tood">TOOD (ICCV'2021)</a></li>
             <li><a href="configs/ddod">DDOD (ACM MM'2021)</a></li>
+            <li><a href="configs/RTMDet">RTMDet (ArXiv'2022)</a></li>
       </ul>
       </td>
       <td>
@@ -207,6 +227,7 @@ MMDetection 是一个基于 PyTorch 的目标检测开源工具箱。它是 [Ope
           <li><a href="configs/mask2former">Mask2Former (ArXiv'2021)</a></li>
           <li><a href="configs/condinst">CondInst (ECCV 2020)</a></li>
           <li><a href="projects/SparseInst">SparseInst (CVPR 2022)</a></li>
+          <li><a href="configs/RTMDet">RTMDet (ArXiv'2022)</a></li>
         </ul>
       </td>
       <td>