PyTorch Implementation for Our Paper: "Object Detection Made Simpler by Eliminating Heuristic NMS"
- Python 3.7
- PyTorch 1.5.1
- mmdetectoin
The code is being submitted to the company for open source review.
| Model | Backbone | MS Training | lr sched | mAP (COCO2017 val) | link |
|---|---|---|---|---|---|
| FCOS | R50 | Yes | 1x | 38.6 | soon |
| FCOS | R50 | Yes | 2x | 41.0 | soon |
| FCOS | R50 | Yes | 3x | 42.0 | soon |
| ATSS | R50 | Yes | 1x | 39.5 | soon |
| ATSS | R50 | Yes | 2x | 41.9 | soon |
| ATSS | R50 | Yes | 3x | 42.8 | soon |
| FCOSPss | R50 | Yes | 3x | 42.3 | soon |
| ATSSPss | R50 | Yes | 3x | 42.6 | soon |
| ATSSPss | R101 | Yes | 3x | 44.2 | |
| ATSSPss | X-101-32x4d-DCN | Yes | 3x | 47.5 | |
| ATSSPss | R2N-101-DCN | Yes | 3x | 48.5 |
If we have a pretrained model, only finetuning the PSS head can save the training time.
| Model | Backbone | MS Training | lr sched | mAP pretrain model (w NMS) |
mAP finetuned PSS (w/o NMS) |
|---|---|---|---|---|---|
| GFocalV2Pss | R50 | Yes | 12 | 43.9 | 43.3 |
| GFocalV2Pss | X-101-32x4d-DCN | Yes | 12 | 48.8 | 48.2 |
| GFocalV2Pss | R2N-101-DCN | Yes | 12 | 49.9 | 49.2 |
| Model | NMS | Backbone | MS Training | lr sched | bbox mAP (COCO2017 val) | segm mAP (COCO2017 val) | link |
|---|---|---|---|---|---|---|---|
| CondInst | Yes | R50 | Yes | 1x | 38.9 | 34.1 | |
| CondInst | Yes | R50 | Yes | 3x | 42.1 | 37.0 | |
| CondInstPss | No | R50 | Yes | 1x | 39.4 | 34.4 | |
| CondInstPss | No | R50 | Yes | 3x | 42.4 | 36.8 | |
| CondInstPss | No | R101 | Yes | 3x | 44.0 | 38.2 |
If you use the package in your research, please cite our paper:
@misc{zhou2021object,
title={Object Detection Made Simpler by Eliminating Heuristic NMS},
author={Qiang Zhou and Chaohui Yu and Chunhua Shen and Zhibin Wang and Hao Li},
year={2021},
eprint={2101.11782},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
团队长期招聘中，社招/实习/校招，我们都要 !
欢迎投递简历，邮箱: [email protected]
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent