Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Feature] Support RTMDet-Ins and improve RTMDet test config. #9494

Merged
merged 18 commits into from
Dec 19, 2022
Merged

[Feature] Support RTMDet-Ins and improve RTMDet test config. #9494

Show file tree
Hide file tree
Changes from 17 commits
Commits
Show all changes
18 commits
Select commit Hold shift + click to select a range
4796460
[Feature] Support RTMDet instance segmentation model.
RangiLyu Nov 29, 2022
bc2203f
remove sampler
RangiLyu Nov 29, 2022
091d681
fix batch
RangiLyu Nov 30, 2022
4b7c5c4
fix act
RangiLyu Nov 30, 2022
2f0de6e
fix act
RangiLyu Dec 7, 2022
193d205
[Enhance] Improve RTMDet AP with YOLO test config.
RangiLyu Dec 15, 2022
5bdbeba
Merge branch 'rtmdet_instance' into update_rtmdet
RangiLyu Dec 16, 2022
f63eb18
update
RangiLyu Dec 16, 2022
fd887ee
update docstring
RangiLyu Dec 16, 2022
d2c4f8b
clean code
RangiLyu Dec 16, 2022
5ce5e09
update
RangiLyu Dec 17, 2022
3785af9
update readme
RangiLyu Dec 16, 2022
88c251d
update readme
RangiLyu Dec 19, 2022
3b1410a
release model weights
RangiLyu Dec 19, 2022
7c74f9e
update readme
RangiLyu Dec 19, 2022
ed853b8
update readme
RangiLyu Dec 19, 2022
d012ba1
update readme
RangiLyu Dec 19, 2022
d223850
update readme
RangiLyu Dec 19, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 18 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,6 +75,24 @@ Apart from MMDetection, we also released [MMEngine](https://github.com/open-mmla

RangiLyu marked this conversation as resolved.
Show resolved Hide resolved
## 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 becomes the state-of-the-art on instance segmentation and rotated object detection tasks. Detailed explanation of its methodologies is available in our [technical report](https://arxiv.org/abs/2212.07784). Pre-trained models can be found [here](configs/rtmdet).
RangiLyu marked this conversation as resolved.
Show resolved Hide resolved

[![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 |
ZwwWayne marked this conversation as resolved.
Show resolved Hide resolved
| 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)
Expand Down
18 changes: 18 additions & 0 deletions README_zh-CN.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,6 +74,24 @@ MMDetection 是一个基于 PyTorch 的目标检测开源工具箱。它是 [Ope

## 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。 我们的[技术报告](https://arxiv.org/abs/2212.07784)中提供了其方法的详细说明。 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)
Expand Down
78 changes: 68 additions & 10 deletions configs/rtmdet/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,25 +1,83 @@
# RTMDet
# RTMDet: An Empirical Study of Designing Real-Time Object Detectors

[![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)

<!-- [ALGORITHM] -->

## Abstract

Our tech-report will be released soon.
In this paper, we aim to design an efficient real-time object detector that exceeds the YOLO series and is easily extensible for many object recognition tasks such as instance segmentation and rotated object detection. To obtain a more efficient model architecture, we explore an architecture that has compatible capacities in the backbone and neck, constructed by a basic building block that consists of large-kernel depth-wise convolutions. We further introduce soft labels when calculating matching costs in the dynamic label assignment to improve accuracy. Together with better training techniques, the resulting object detector, named RTMDet, achieves 52.8% AP on COCO with 300+ FPS on an NVIDIA 3090 GPU, outperforming the current mainstream industrial detectors. RTMDet achieves the best parameter-accuracy trade-off with tiny/small/medium/large/extra-large model sizes for various application scenarios, and obtains new state-of-the-art performance on real-time instance segmentation and rotated object detection. We hope the experimental results can provide new insights into designing versatile real-time object detectors for many object recognition tasks.

<div align=center>
<img src="https://user-images.githubusercontent.com/12907710/192182907-f9a671d6-89cb-4d73-abd8-c2b9dada3c66.png"/>
<img src="https://user-images.githubusercontent.com/12907710/208070055-7233a3d8-955f-486a-82da-b714b3c3bbd6.png"/>
</div>

## Results and Models

| Backbone | size | box AP | Params(M) | FLOPS(G) | TRT-FP16-Latency(ms) | Config | Download |
## Object Detection

| Model | size | box AP | Params(M) | FLOPS(G) | TRT-FP16-Latency(ms) | Config | Download |
| :---------: | :--: | :----: | :-------: | :------: | :------------------: | :----------------------------------------: | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| RTMDet-tiny | 640 | 40.9 | 4.8 | 8.1 | 0.98 | [config](./rtmdet_tiny_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_tiny_8xb32-300e_coco/rtmdet_tiny_8xb32-300e_coco_20220902_112414-78e30dcc.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_tiny_8xb32-300e_coco/rtmdet_tiny_8xb32-300e_coco_20220902_112414.log.json) |
| RTMDet-s | 640 | 44.5 | 8.89 | 14.8 | 1.22 | [config](./rtmdet_s_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_s_8xb32-300e_coco/rtmdet_s_8xb32-300e_coco_20220905_161602-387a891e.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_s_8xb32-300e_coco/rtmdet_s_8xb32-300e_coco_20220905_161602.log.json) |
| RTMDet-m | 640 | 49.1 | 24.71 | 39.27 | 1.62 | [config](./rtmdet_m_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_m_8xb32-300e_coco/rtmdet_m_8xb32-300e_coco_20220719_112220-229f527c.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_m_8xb32-300e_coco/rtmdet_m_8xb32-300e_coco_20220719_112220.log.json) |
| RTMDet-l | 640 | 51.3 | 52.3 | 80.23 | 2.44 | [config](./rtmdet_l_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_l_8xb32-300e_coco/rtmdet_l_8xb32-300e_coco_20220719_112030-5a0be7c4.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_l_8xb32-300e_coco/rtmdet_l_8xb32-300e_coco_20220719_112030.log.json) |
| RTMDet-x | 640 | 52.6 | 94.86 | 141.67 | 3.10 | [config](./rtmdet_x_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_x_8xb32-300e_coco/rtmdet_x_8xb32-300e_coco_20220715_230555-cc79b9ae.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_x_8xb32-300e_coco/rtmdet_x_8xb32-300e_coco_20220715_230555.log.json) |
| RTMDet-tiny | 640 | 41.1 | 4.8 | 8.1 | 0.98 | [config](./rtmdet_tiny_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_tiny_8xb32-300e_coco/rtmdet_tiny_8xb32-300e_coco_20220902_112414-78e30dcc.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_tiny_8xb32-300e_coco/rtmdet_tiny_8xb32-300e_coco_20220902_112414.log.json) |
| RTMDet-s | 640 | 44.6 | 8.89 | 14.8 | 1.22 | [config](./rtmdet_s_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_s_8xb32-300e_coco/rtmdet_s_8xb32-300e_coco_20220905_161602-387a891e.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_s_8xb32-300e_coco/rtmdet_s_8xb32-300e_coco_20220905_161602.log.json) |
| RTMDet-m | 640 | 49.4 | 24.71 | 39.27 | 1.62 | [config](./rtmdet_m_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_m_8xb32-300e_coco/rtmdet_m_8xb32-300e_coco_20220719_112220-229f527c.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_m_8xb32-300e_coco/rtmdet_m_8xb32-300e_coco_20220719_112220.log.json) |
| RTMDet-l | 640 | 51.5 | 52.3 | 80.23 | 2.44 | [config](./rtmdet_l_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_l_8xb32-300e_coco/rtmdet_l_8xb32-300e_coco_20220719_112030-5a0be7c4.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_l_8xb32-300e_coco/rtmdet_l_8xb32-300e_coco_20220719_112030.log.json) |
| RTMDet-x | 640 | 52.8 | 94.86 | 141.67 | 3.10 | [config](./rtmdet_x_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_x_8xb32-300e_coco/rtmdet_x_8xb32-300e_coco_20220715_230555-cc79b9ae.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet_x_8xb32-300e_coco/rtmdet_x_8xb32-300e_coco_20220715_230555.log.json) |

**Note**:

1. The inference speed is measured on an NVIDIA 3090 GPU with TensorRT 8.4.3, cuDNN 8.2.0, FP16, batch size=1, and without NMS.
1. The inference speed of RTMDet is measured on an NVIDIA 3090 GPU with TensorRT 8.4.3, cuDNN 8.2.0, FP16, batch size=1, and without NMS.
2. For a fair comparison, the config of bbox postprocessing is changed to be consistent with YOLOv5/6/7 after [PR#9494](https://github.com/open-mmlab/mmdetection/pull/9494), bringing about 0.1~0.3% AP improvement.

## Instance Segmentation

RTMDet-Ins is the state-of-the-art real-time instance segmentation on coco dataset:

[![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)

| Model | size | box AP | mask AP | Params(M) | FLOPS(G) | TRT-FP16-Latency(ms) | Config | Download |
| :-------------: | :--: | :----: | :-----: | :-------: | :------: | :------------------: | :--------------------------------------------: | :--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| RTMDet-Ins-tiny | 640 | 40.5 | 35.4 | 5.6 | 11.8 | 1.70 | [config](./rtmdet-ins_tiny_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_tiny_8xb32-300e_coco/rtmdet-ins_tiny_8xb32-300e_coco_20221130_151727-ec670f7e.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_tiny_8xb32-300e_coco/rtmdet-ins_tiny_8xb32-300e_coco_20221130_151727.log.json) |
| RTMDet-Ins-s | 640 | 44.0 | 38.7 | 10.18 | 21.5 | 1.93 | [config](./rtmdet-ins_s_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_s_8xb32-300e_coco/rtmdet-ins_s_8xb32-300e_coco_20221121_212604-fdc5d7ec.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_s_8xb32-300e_coco/rtmdet-ins_s_8xb32-300e_coco_20221121_212604.log.json) |
| RTMDet-Ins-m | 640 | 48.8 | 42.1 | 27.58 | 54.13 | 2.69 | [config](./rtmdet-ins_m_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_m_8xb32-300e_coco/rtmdet-ins_m_8xb32-300e_coco_20221123_001039-6eba602e.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_m_8xb32-300e_coco/rtmdet-ins_m_8xb32-300e_coco_20221123_001039.log.json) |
| RTMDet-Ins-l | 640 | 51.2 | 43.7 | 57.37 | 106.56 | 3.68 | [config](./rtmdet-ins_l_8xb32-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_l_8xb32-300e_coco/rtmdet-ins_l_8xb32-300e_coco_20221124_103237-78d1d652.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_l_8xb32-300e_coco/rtmdet-ins_l_8xb32-300e_coco_20221124_103237.log.json) |
| RTMDet-Ins-x | 640 | 52.4 | 44.6 | 102.7 | 182.7 | 5.31 | [config](./rtmdet-ins_x_8xb16-300e_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_x_8xb16-300e_coco/rtmdet-ins_x_8xb16-300e_coco_20221124_111313-33d4595b.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/rtmdet/rtmdet-ins_l_8xb16-300e_coco/rtmdet-ins_x_8xb16-300e_coco_20221124_111313.log.json) |

**Note**:

1. The inference speed of RTMDet-Ins is measured on an NVIDIA 3090 GPU with TensorRT 8.4.3, cuDNN 8.2.0, FP16, batch size=1. Top 100 masks are kept and the post process latency is included.

## Rotated Object Detection

RTMDet-R achieves state-of-the-art on various remote sensing datasets

[![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/one-stage-anchor-free-oriented-object-1)](https://paperswithcode.com/sota/one-stage-anchor-free-oriented-object-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)

[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/rtmdet-an-empirical-study-of-designing-real/one-stage-anchor-free-oriented-object-3)](https://paperswithcode.com/sota/one-stage-anchor-free-oriented-object-3?p=rtmdet-an-empirical-study-of-designing-real)

Models and configs of RTMDet-R are available in [MMRotate](https://github.com/open-mmlab/mmrotate/tree/1.x/configs/rotated_rtmdet)

## Citation

```latex
@misc{lyu2022rtmdet,
title={RTMDet: An Empirical Study of Designing Real-Time Object Detectors},
author={Chengqi Lyu and Wenwei Zhang and Haian Huang and Yue Zhou and Yudong Wang and Yanyi Liu and Shilong Zhang and Kai Chen},
year={2022},
eprint={2212.07784},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```

## Visualization

<div align=center>
<img src="https://user-images.githubusercontent.com/12907710/208044554-1e8de6b5-48d8-44e4-a7b5-75076c7ebb71.png"/>
</div>
Loading