arXiv | Project Page | Code
This repository contains the implementation of the following paper:
Open-Vocabulary DETR with Conditional Matching
Yuhang Zang, Wei Li, Kaiyang Zhou, Chen Huang, Chen Change Loy
European Conference on Computer Vision (ECCV), 2022
We use the same environment as Deformable DETR. You are also required to install the following packages:
We test our models under python=3.8, pytorch=1.11.0, cuda=10.1, 8 Nvidia V100 32GB GPUs.
Please refer to dataset_prepare.md.
Please refer to run_scripts.md.
- Open-vocabulary COCO (AP50 metric)
| Base | Novel | All | Model |
|---|---|---|---|
| 61.0 | 29.4 | 52.7 | Google Drive |
If you find our work useful for your research, please consider citing the paper:
@InProceedings{zang2022open,
author = {Zang, Yuhang and Li, Wei and Zhou, Kaiyang and Huang, Chen and Loy, Chen Change},
title = {Open-Vocabulary DETR with Conditional Matching},
journal = {European Conference on Computer Vision},
year = {2022}
}
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
We would like to thanks Deformable DETR, CLIP and ViLD for their open-source projects.
Please contact Yuhang Zang if you have any questions.