README.md 3.1 KB
Newer Older
G
Guanghua Yu 已提交
1 2 3 4 5 6
# SOLOv2 for instance segmentation

## Introduction

SOLOv2 (Segmenting Objects by Locations) is a fast instance segmentation framework with strong performance. We reproduced the model of the paper, and improved and optimized the accuracy and speed of the SOLOv2.

G
Guanghua Yu 已提交
7 8
**Highlights:**

G
Guanghua Yu 已提交
9 10 11
- Performance: `Light-R50-VD-DCN-FPN` model reached 38.6 FPS on single Tesla V100, and mask ap on the COCO-val dataset reached 38.8, which increased inference speed by 24%, mAP increased by 2.4 percentage points.
- Training Time: The training time of the model of `solov2_r50_fpn_1x` on Tesla v100 with 8 GPU is only 10 hours.

12 13 14 15
<div align="center">
  <img src="../../docs/images/instance_segmentation.png" width=800 />
</div>

G
Guanghua Yu 已提交
16 17 18

## Model Zoo

19
| Backbone                | Multi-scale training  | Lr schd | V100 FP32(FPS) | Mask AP<sup>val</sup> |         Download                  | Configs |
G
Guanghua Yu 已提交
20
| :---------------------: | :-------------------: | :-----: | :------------: | :-----: | :---------: | :------------------------: |
21 22 23 24
| Mobilenetv3-FPN                 |  True                |   3x    |     50          |  30.0   | [model](https://paddlemodels.bj.bcebos.com/object_detection/solov2_mobilenetv3_fpn_448_3x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/solov2/solov2_mobilenetv3_fpn_448_3x.yml) |
| R50-FPN                 |  False                |   1x    |     21.9          |  35.6   | [model](https://paddlemodels.bj.bcebos.com/object_detection/solov2_r50_fpn_1x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/solov2/solov2_r50_fpn_1x.yml) |
| R50-FPN                 |  True                |   3x    |     21.9          |  37.9   | [model](https://paddlemodels.bj.bcebos.com/object_detection/solov2_r50_fpn_3x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/solov2/solov2_r50_fpn_3x.yml) |
| R101-VD-FPN                 |  True               |   3x    |     12.1       |  42.6   | [model](https://paddlemodels.bj.bcebos.com/object_detection/solov2_r101_vd_fpn_3x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/solov2/solov2_r101_vd_fpn_3x.yml) |
G
Guanghua Yu 已提交
25 26

## Enhanced model
27
| Backbone                | Input size  | Lr schd | V100 FP32(FPS) | Mask AP<sup>val</sup> |         Download                  | Configs |
G
Guanghua Yu 已提交
28
| :---------------------: | :-------------------: | :-----: | :------------: | :-----: | :---------: | :------------------------: |
29
| Light-R50-VD-DCN-FPN          |  512     |   3x    |     38.6          |  38.8   | [model](https://paddlemodels.bj.bcebos.com/object_detection/solov2_light_r50_vd_fpn_dcn_512_3x.pdparams) | [config](https://github.com/PaddlePaddle/PaddleDetection/tree/master/configs/solov2/solov2_light_r50_vd_fpn_dcn_512_3x.yml) |
G
Guanghua Yu 已提交
30

31 32 33 34
**Notes:**

- SOLOv2 is trained on COCO train2017 dataset and evaluated on val2017 results of `mAP(IoU=0.5:0.95)`.

G
Guanghua Yu 已提交
35 36 37 38 39 40 41 42 43
## Citations
```
@article{wang2020solov2,
  title={SOLOv2: Dynamic, Faster and Stronger},
  author={Wang, Xinlong and Zhang, Rufeng and  Kong, Tao and Li, Lei and Shen, Chunhua},
  journal={arXiv preprint arXiv:2003.10152},
  year={2020}
}
```