https://github.com/steven-lang/DAFNe
Code for our paper "DAFNe: A One-Stage Anchor-Free Deep Model for Oriented Object Detection".
https://github.com/steven-lang/DAFNe
anchor-free deep-learning machine-learning object-detection one-stage-detector oriented-object-detection
Last synced: 3 months ago
JSON representation
Code for our paper "DAFNe: A One-Stage Anchor-Free Deep Model for Oriented Object Detection".
- Host: GitHub
- URL: https://github.com/steven-lang/DAFNe
- Owner: braun-steven
- License: mit
- Created: 2021-09-10T05:22:04.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2022-04-01T05:50:11.000Z (about 3 years ago)
- Last Synced: 2024-08-05T01:07:26.260Z (8 months ago)
- Topics: anchor-free, deep-learning, machine-learning, object-detection, one-stage-detector, oriented-object-detection
- Language: Python
- Homepage:
- Size: 2.99 MB
- Stars: 60
- Watchers: 1
- Forks: 12
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- awesome-anchor-free-object-detection - DAFNe - lang/DAFNe?style=social"/> : "DAFNe: A One-Stage Anchor-Free Deep Model for Oriented Object Detection". (**[arXiv 2021](https://arxiv.org/abs/2109.06148)**) (Frameworks)
README
# DAFNe: A One-Stage Anchor-Free Approach for Oriented Object Detection
Code for our Paper [DAFNe: A One-Stage Anchor-Free Approach for Oriented Object Detection](https://arxiv.org/abs/2109.06148).
[](https://paperswithcode.com/sota/one-stage-anchor-free-oriented-object-1?p=dafne-a-one-stage-anchor-free-deep-model-for)
[](https://paperswithcode.com/sota/one-stage-anchor-free-oriented-object-2?p=dafne-a-one-stage-anchor-free-deep-model-for)
[](https://paperswithcode.com/sota/one-stage-anchor-free-oriented-object-3?p=dafne-a-one-stage-anchor-free-deep-model-for)
## Datasets
- UCAS-AOD: https://hyper.ai/datasets/5419
- DOTA 1.0/1.5: https://captain-whu.github.io/DOTA/index.html
- Note: See [./tools/prepare_dota/](./tools/prepare_dota/) for instructions on how to prepare the DOTA datasets.
- HRSC2016: https://www.kaggle.com/guofeng/hrsc2016
## Docker Setup
Use the `Dockerfile` to build the necessary docker image:
``` bash
docker build -t dafne .
```
## Training
Check out `./configs/pre-trained/` for different pre-defined configurations for the DOTA 1.0, DOTA 1.5, UCAS-AOD, and HRSC2016 datasets. Use these paths as argument for the `--config-file` option below.
### With Docker
Use the `./tools/run.py` helper to start running experiments
``` bash
./tools/run.py --gpus 0,1,2,3 --config-file ./configs/dota-1.0/1024.yaml
```
### Without Docker
``` bash
NVIDIA_VISIBLE_DEVICES=0,1,2,3 ./tools/plain_train_net.py --num-gpus 4 --config-file ./configs/dota-1.0/1024.yaml
```
## Pre-Trained Weights
| Dataset | mAP (%) | Config | Weights |
|----------|---------|-----------------------------------------------------------------|------------------------------------------------------------------------------------------------------------|
| UCAS-AOD | 89.65 | [ucas_aod_r101_ms](./configs/pre-trained/ucas_aod_r101_ms.yaml) | [ucas-aod-r101-ms.pth](https://drive.google.com/file/d/1snC7IU-ud-d6L_AxbDx_HG8QBINP2_RO/view?usp=sharing) |
| HRSC2016 | 89.76 | [hrsc_r50_ms](./configs/pre-trained/hrsc_r50_ms.yaml) | [hrsc-r50-ms.pth](https://drive.google.com/file/d/10i3pHxiHgjJGzJoZK-HtNdsAyfGD5Ydj/view?usp=sharing) |
| DOTA 1.0 | 76.95 | [dota-1.0_r101_ms](./configs/pre-trained/dota-1.0_r101_ms.yaml) | [dota-1.0-r101-ms.pth](https://drive.google.com/file/d/1-lgSLhKQSZBogI2YD0r64wjJV6k2xL4E/view?usp=sharing) |
| DOTA 1.5 | 71.99 | [dota-1.5_r101_ms](./configs/pre-trained/dota-1.5_r101_ms.yaml) | [dota-1.5-r101-ms.pth](https://drive.google.com/file/d/1MQbTngieoWh-DcJL-z55RnI3PUNeSvBv/view?usp=sharing) |
### Pre-Trained Weights Usage with Docker
``` bash
./tools/run.py --gpus 0 --config-file --opts "MODEL.WEIGHTS "
```
### Pre-Trained Weights Usage without Docker
``` bash
NVIDIA_VISIBLE_DEVICES=0 ./tools/plain_train_net.py --num-gpus 1 --config-file MODEL.WEIGHTS
```
## Cite
``` bibtex
@misc{lang2021dafne,
title={DAFNe: A One-Stage Anchor-Free Approach for Oriented Object Detection},
author={Steven Lang and Fabrizio Ventola and Kristian Kersting},
year={2021},
eprint={2109.06148},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
## Acknowledgments
- Thanks to [AdelaiDet](https://github.com/aim-uofa/AdelaiDet) for providing the initial FCOS implementation
- Thanks to [Detectron2](https://github.com/facebookresearch/detectron2) for providing a general object detection framework