Skip to content

Commit

Permalink
Update docs
Browse files Browse the repository at this point in the history
  • Loading branch information
shinya7y committed Oct 12, 2022
1 parent 8247e20 commit 321a285
Show file tree
Hide file tree
Showing 7 changed files with 39 additions and 27 deletions.
6 changes: 3 additions & 3 deletions CITATION.cff
Original file line number Diff line number Diff line change
Expand Up @@ -8,10 +8,10 @@ date-released: 2020-06-26
url: "https://github.com/shinya7y/UniverseNet"
license: Apache-2.0
preferred-citation:
type: article
type: conference-paper
authors:
- family-names: "Shinya"
given-names: "Yosuke"
journal: "arXiv:2103.14027"
collection-title: "British Machine Vision Conference (BMVC)"
title: "USB: Universal-Scale Object Detection Benchmark"
year: 2021
year: 2022
31 changes: 20 additions & 11 deletions README.md
Original file line number Diff line number Diff line change
@@ -1,14 +1,23 @@
# UniverseNet

UniverseNets are state-of-the-art detectors for universal-scale object detection.
Please refer to our paper for details.
https://arxiv.org/abs/2103.14027
This is the official repository of "[USB: Universal-Scale Object Detection Benchmark](https://arxiv.org/abs/2103.14027)" (BMVC 2022).

We established a new benchmark *USB* with fair protocols and designed state-of-the-art detectors *UniverseNets* for universal-scale object detection.
This repository extends MMDetection with more features and allows for more comprehensive benchmarking and development.

## Introduction

![universal-scale object detection](https://user-images.githubusercontent.com/42844407/113513063-b5aa2780-95a2-11eb-8413-2fb470256a1a.png)
![COCO AP](https://user-images.githubusercontent.com/42844407/112571346-317dd480-8e2b-11eb-8b1e-a97ba2e97997.png)

Benchmarks, such as COCO, play a crucial role in object detection. However, existing benchmarks are insufficient in scale variation, and their protocols are inadequate for fair comparison. In this paper, we introduce the Universal-Scale object detection Benchmark (USB). USB has variations in object scales and image domains by incorporating COCO with the recently proposed Waymo Open Dataset and Manga109-s dataset. To enable fair comparison and inclusive research, we propose training and evaluation protocols. They have multiple divisions for training epochs and evaluation image resolutions, like weight classes in sports, and compatibility across training protocols, like the backward compatibility of the Universal Serial Bus. Specifically, we request participants to report results with not only higher protocols (longer training) but also lower protocols (shorter training). Using the proposed benchmark and protocols, we conducted extensive experiments using 15 methods and found weaknesses of existing COCO-biased methods.

Please refer to our paper for details.
https://arxiv.org/abs/2103.14027

## Changelog

- recent:
- Our paper got accepted to BMVC 2022!
- 22.06 (June 2022):
- Add SwinV2, FocalNet, GBR COTS dataset
- Update codes for mmdet 2.25.0, mmcv-full 1.4.4
Expand Down Expand Up @@ -45,14 +54,14 @@ https://arxiv.org/abs/2103.14027

Methods and architectures:

- [x] [UniverseNets (arXiv 2021)](configs/universenet/)
- [x] [UniverseNets (BMVC 2022)](configs/universenet/)
- [x] [FocalNet (NeurIPS 2022)](configs/focalnet/)
- [x] [Swin Transformer V2 (CVPR 2022)](configs/swinv2/)
- [x] [FocalNet (arXiv 2022)](configs/focalnet/)
- [x] [PoolFormer (CVPR 2022)](configs/poolformer/)
- [x] [PVTv2 (CVMJ 2022)](configs/pvtv2_original/) stronger models
- [x] [ConvMLP (arXiv 2021)](configs/convmlp/)
- [x] [CBNetV2 (arXiv 2021)](configs/cbnet/)
- [x] ~~TOOD (ICCV 2021)~~ supported
- [x] [PVTv2 (CVMJ 2022)](configs/pvtv2_original/) stronger models
- [x] ~~PVT (ICCV 2021)~~ supported
- [x] [Swin Transformer (ICCV 2021)](configs/swin_original/) stronger models
- [x] ~~DDOD (ACMMM 2021)~~ supported
Expand All @@ -64,7 +73,7 @@ Methods and architectures:

Benchmarks and datasets:

- [x] USB (arXiv 2021)
- [x] USB (BMVC 2022)
- [x] [Waymo Open Dataset (CVPR 2020)](configs/waymo_open/)
- [x] [Manga109-s dataset (MTAP 2017, IEEE MultiMedia 2020)](configs/manga109/)
- [x] [NightOwls dataset (ACCV 2018)](configs/nightowls/)
Expand Down Expand Up @@ -107,11 +116,11 @@ Even if you have one GPU, we recommend using `tools/dist_train.sh` and `tools/di
## Citation

```
@article{USB_shinya_2021,
@inproceedings{USB_shinya_BMVC2022,
title={{USB}: Universal-Scale Object Detection Benchmark},
author={Shinya, Yosuke},
journal={arXiv:2103.14027},
year={2021}
booktitle={British Machine Vision Conference (BMVC)},
year={2022}
}
```

Expand Down
6 changes: 3 additions & 3 deletions configs/manga109/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -89,10 +89,10 @@ Please cite the following paper for the benchmark results.
https://arxiv.org/abs/2103.14027

```
@article{USB_shinya_2021,
@inproceedings{USB_shinya_BMVC2022,
title={{USB}: Universal-Scale Object Detection Benchmark},
author={Shinya, Yosuke},
journal={arXiv:2103.14027},
year={2021}
booktitle={British Machine Vision Conference (BMVC)},
year={2022}
}
```
5 changes: 4 additions & 1 deletion configs/nightowls/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -62,11 +62,14 @@ UniverseNet achieves the 1st place in the NightOwls Detection Challenge 2020 all
}
```

For the results of UniverseNet, please cite the arXiv v2 version of our paper.
https://arxiv.org/abs/2103.14027v2

```
@article{USB_shinya_2021,
title={{USB}: Universal-Scale Object Detection Benchmark},
author={Shinya, Yosuke},
journal={arXiv:2103.14027},
journal={arXiv:2103.14027v2},
year={2021}
}
```
6 changes: 3 additions & 3 deletions configs/universenet/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -9,11 +9,11 @@ Please refer to our paper for details.
https://arxiv.org/abs/2103.14027

```
@article{USB_shinya_2021,
@inproceedings{USB_shinya_BMVC2022,
title={{USB}: Universal-Scale Object Detection Benchmark},
author={Shinya, Yosuke},
journal={arXiv:2103.14027},
year={2021}
booktitle={British Machine Vision Conference (BMVC)},
year={2022}
}
```

Expand Down
6 changes: 3 additions & 3 deletions configs/waymo_open/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -135,10 +135,10 @@ Please cite the following paper for the benchmark results.
https://arxiv.org/abs/2103.14027

```
@article{USB_shinya_2021,
@inproceedings{USB_shinya_BMVC2022,
title={{USB}: Universal-Scale Object Detection Benchmark},
author={Shinya, Yosuke},
journal={arXiv:2103.14027},
year={2021}
booktitle={British Machine Vision Conference (BMVC)},
year={2022}
}
```
6 changes: 3 additions & 3 deletions docs/en/tutorials/finer_scale_ap.md
Original file line number Diff line number Diff line change
Expand Up @@ -51,10 +51,10 @@ Please cite the following paper.
https://arxiv.org/abs/2103.14027

```
@article{USB_shinya_2021,
@inproceedings{USB_shinya_BMVC2022,
title={{USB}: Universal-Scale Object Detection Benchmark},
author={Shinya, Yosuke},
journal={arXiv:2103.14027},
year={2021}
booktitle={British Machine Vision Conference (BMVC)},
year={2022}
}
```

0 comments on commit 321a285

Please sign in to comment.