File size: 11,513 Bytes
186701e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 |
# YOLOX
> [YOLOX: Exceeding YOLO Series in 2021](https://arxiv.org/abs/2107.08430)
<!-- [ALGORITHM] -->
## Abstract
In this report, we present some experienced improvements to YOLO series, forming a new high-performance detector -- YOLOX. We switch the YOLO detector to an anchor-free manner and conduct other advanced detection techniques, i.e., a decoupled head and the leading label assignment strategy SimOTA to achieve state-of-the-art results across a large scale range of models: For YOLO-Nano with only 0.91M parameters and 1.08G FLOPs, we get 25.3% AP on COCO, surpassing NanoDet by 1.8% AP; for YOLOv3, one of the most widely used detectors in industry, we boost it to 47.3% AP on COCO, outperforming the current best practice by 3.0% AP; for YOLOX-L with roughly the same amount of parameters as YOLOv4-CSP, YOLOv5-L, we achieve 50.0% AP on COCO at a speed of 68.9 FPS on Tesla V100, exceeding YOLOv5-L by 1.8% AP. Further, we won the 1st Place on Streaming Perception Challenge (Workshop on Autonomous Driving at CVPR 2021) using a single YOLOX-L model. We hope this report can provide useful experience for developers and researchers in practical scenes, and we also provide deploy versions with ONNX, TensorRT, NCNN, and Openvino supported.
<div align=center>
<img src="https://user-images.githubusercontent.com/40661020/144001736-9fb303dd-eac7-46b0-ad45-214cfa51e928.png"/>
</div>
<div align=center>
<img src="https://user-images.githubusercontent.com/71306851/218628641-6c0101e6-e40e-4b16-a696-c0f55b8d335c.png"/>
YOLOX-l model structure
</div>
## 🥳 🚀 Results and Models
| Backbone | Size | Batch Size | AMP | RTMDet-Hyp | Mem (GB) | Box AP | Config | Download |
| :--------: | :--: | :--------: | :-: | :--------: | :------: | :---------: | :-------------------------------------------------------: | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| YOLOX-tiny | 416 | 8xb8 | No | No | 2.8 | 32.7 | [config](./yolox_tiny_fast_8xb8-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_tiny_8xb8-300e_coco/yolox_tiny_8xb8-300e_coco_20220919_090908-0e40a6fc.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_tiny_8xb8-300e_coco/yolox_tiny_8xb8-300e_coco_20220919_090908.log.json) |
| YOLOX-tiny | 416 | 8xb32 | Yes | Yes | 4.9 | 34.3 (+1.6) | [config](./yolox_tiny_fast_8xb32-300e-rtmdet-hyp_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_tiny_fast_8xb32-300e-rtmdet-hyp_coco/yolox_tiny_fast_8xb32-300e-rtmdet-hyp_coco_20230210_143637-4c338102.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_tiny_fast_8xb32-300e-rtmdet-hyp_coco/yolox_tiny_fast_8xb32-300e-rtmdet-hyp_coco_20230210_143637.log.json) |
| YOLOX-s | 640 | 8xb8 | Yes | No | 2.9 | 40.7 | [config](./yolox_s_fast_8xb8-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_s_fast_8xb8-300e_coco/yolox_s_fast_8xb8-300e_coco_20230213_142600-2b224d8b.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_s_fast_8xb8-300e_coco/yolox_s_fast_8xb8-300e_coco_20230213_142600.log.json) |
| YOLOX-s | 640 | 8xb32 | Yes | Yes | 9.8 | 41.9 (+1.2) | [config](./yolox_s_fast_8xb32-300e-rtmdet-hyp_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_s_fast_8xb32-300e-rtmdet-hyp_coco/yolox_s_fast_8xb32-300e-rtmdet-hyp_coco_20230210_134645-3a8dfbd7.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_s_fast_8xb32-300e-rtmdet-hyp_coco/yolox_s_fast_8xb32-300e-rtmdet-hyp_coco_20230210_134645.log.json) |
| YOLOX-m | 640 | 8xb8 | Yes | No | 4.9 | 46.9 | [config](./yolox_m_fast_8xb8-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_m_fast_8xb8-300e_coco/yolox_m_fast_8xb8-300e_coco_20230213_160218-a71a6b25.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_m_fast_8xb8-300e_coco/yolox_m_fast_8xb8-300e_coco_20230213_160218.log.json) |
| YOLOX-m | 640 | 8xb32 | Yes | Yes | 17.6 | 47.5 (+0.6) | [config](./yolox_m_fast_8xb32-300e-rtmdet-hyp_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_m_fast_8xb32-300e-rtmdet-hyp_coco/yolox_m_fast_8xb32-300e-rtmdet-hyp_coco_20230210_144328-e657e182.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_m_fast_8xb32-300e-rtmdet-hyp_coco/yolox_m_fast_8xb32-300e-rtmdet-hyp_coco_20230210_144328.log.json) |
| YOLOX-l | 640 | 8xb8 | Yes | No | 8.0 | 50.1 | [config](./yolox_l_fast_8xb8-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_l_fast_8xb8-300e_coco/yolox_l_fast_8xb8-300e_coco_20230213_160715-c731eb1c.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_l_fast_8xb8-300e_coco/yolox_l_fast_8xb8-300e_coco_20230213_160715.log.json) |
| YOLOX-x | 640 | 8xb8 | Yes | No | 9.8 | 51.4 | [config](./yolox_x_fast_8xb8-300e_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_x_fast_8xb8-300e_coco/yolox_x_fast_8xb8-300e_coco_20230215_133950-1d509fab.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/yolox_x_fast_8xb8-300e_coco/yolox_x_fast_8xb8-300e_coco_20230215_133950.log.json) |
YOLOX uses a default training configuration of `8xbs8` which results in a long training time, we expect it to use `8xbs32` to speed up the training and not cause a decrease in mAP. We modified `train_batch_size_per_gpu` from 8 to 32, `batch_augments_interval` from 10 to 1 and `base_lr` from 0.01 to 0.04 under YOLOX-s default configuration based on the linear scaling rule, which resulted in mAP degradation. Finally, I found that using RTMDet's training hyperparameter can improve performance in YOLOX Tiny/S/M, which also validates the superiority of RTMDet's training hyperparameter.
The modified training parameters are as follows：
1. train_batch_size_per_gpu: 8 -> 32
2. batch_augments_interval: 10 -> 1
3. num_last_epochs: 15 -> 20
4. optim cfg: SGD -> AdamW, base_lr 0.01 -> 0.004, weight_decay 0.0005 -> 0.05
5. ema momentum: 0.0001 -> 0.0002
**Note**:
1. The test score threshold is 0.001.
2. Due to the need for pre-training weights, we cannot reproduce the performance of the `yolox-nano` model. Please refer to https://github.com/Megvii-BaseDetection/YOLOX/issues/674 for more information.
## YOLOX-Pose
Based on [MMPose](https://github.com/open-mmlab/mmpose/blob/main/projects/yolox-pose/README.md), we have implemented a YOLOX-based human pose estimator, utilizing the approach outlined in **YOLO-Pose: Enhancing YOLO for Multi Person Pose Estimation Using Object Keypoint Similarity Loss (CVPRW 2022)**. This pose estimator is lightweight and quick, making it well-suited for crowded scenes.
<div align=center>
<img src="https://user-images.githubusercontent.com/26127467/226655503-3cee746e-6e42-40be-82ae-6e7cae2a4c7e.jpg"/>
</div>
### Results
| Backbone | Size | Batch Size | AMP | RTMDet-Hyp | Mem (GB) | AP | Config | Download |
| :--------: | :--: | :--------: | :-: | :--------: | :------: | :--: | :------------------------------------------------------------: | :----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| YOLOX-tiny | 416 | 8xb32 | Yes | Yes | 5.3 | 52.8 | [config](./pose/yolox-pose_tiny_8xb32-300e-rtmdet-hyp_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/pose/yolox-pose_tiny_8xb32-300e-rtmdet-hyp_coco/yolox-pose_tiny_8xb32-300e-rtmdet-hyp_coco_20230427_080351-2117af67.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/pose/yolox-pose_tiny_8xb32-300e-rtmdet-hyp_coco/yolox-pose_tiny_8xb32-300e-rtmdet-hyp_coco_20230427_080351.log.json) |
| YOLOX-s | 640 | 8xb32 | Yes | Yes | 10.7 | 63.7 | [config](./pose/yolox-pose_s_8xb32-300e-rtmdet-hyp_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/pose/yolox-pose_s_8xb32-300e-rtmdet-hyp_coco/yolox-pose_s_8xb32-300e-rtmdet-hyp_coco_20230427_005150-e87d843a.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/pose/yolox-pose_s_8xb32-300e-rtmdet-hyp_coco/yolox-pose_s_8xb32-300e-rtmdet-hyp_coco_20230427_005150.log.json) |
| YOLOX-m | 640 | 8xb32 | Yes | Yes | 19.2 | 69.3 | [config](./pose/yolox-pose_m_8xb32-300e-rtmdet-hyp_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/pose/yolox-pose_m_8xb32-300e-rtmdet-hyp_coco/yolox-pose_m_8xb32-300e-rtmdet-hyp_coco_20230427_094024-bbeacc1c.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/pose/yolox-pose_m_8xb32-300e-rtmdet-hyp_coco/yolox-pose_m_8xb32-300e-rtmdet-hyp_coco_20230427_094024.log.json) |
| YOLOX-l | 640 | 8xb32 | Yes | Yes | 30.3 | 71.1 | [config](./pose/yolox-pose_l_8xb32-300e-rtmdet-hyp_coco.py) | [model](https://download.openmmlab.com/mmyolo/v0/yolox/pose/yolox-pose_l_8xb32-300e-rtmdet-hyp_coco/yolox-pose_l_8xb32-300e-rtmdet-hyp_coco_20230427_041140-82d65ac8.pth) \| [log](https://download.openmmlab.com/mmyolo/v0/yolox/pose/yolox-pose_l_8xb32-300e-rtmdet-hyp_coco/yolox-pose_l_8xb32-300e-rtmdet-hyp_coco_20230427_041140.log.json) |
**Note**
1. The performance is unstable and may fluctuate and the highest performance weight in `COCO` training may not be the last epoch. The performance shown above is the best model.
### Installation
Install MMPose
```
mim install -r requirements/mmpose.txt
```
## Citation
```latex
@article{yolox2021,
title={{YOLOX}: Exceeding YOLO Series in 2021},
author={Ge, Zheng and Liu, Songtao and Wang, Feng and Li, Zeming and Sun, Jian},
journal={arXiv preprint arXiv:2107.08430},
year={2021}
}
```
|