Official project page of TransMatting (ECCV2022)
The extended paper TransMatting: Tri-token Equipped Transformer Model for Image Matting is published on arXiv. We further propose TransMattingV2, which extends tri-token to both convolutional networks and Transformer. Various experiments demonstrate that it could boost both Transformer and convolutional networks, indicating that the proposed tri-token has the potential to be a new paradigm for image matting.
Quantitative Results on Composition-1k
| Model | SAD | MSE | Grad | Conn | Results Download |
|---|---|---|---|---|---|
| TransMattingV1 | 24.96 | 4.58 | 9.72 | 20.16 | GoogleDrive |
The Transparent-460 dataset is now open to the public. To support the community, we have simplified the access process for researchers. Please follow the guidelines below based on your intended use:
If you use this dataset for non-commercial research, you can download it directly. By downloading, you agree to the Terms & Conditions and MUST cite our paper in your work.
- Download Link: Transparent-460 (Google Drive)
- Requirement: Please include the Citation provided below in any publication or project using this data.
If you intend to use this dataset for any commercial activity, you MUST contact the authors to discuss licensing and permission.
- Contact: Huanqia Cai (caihuanqia19@mails.ucas.ac.cn)
- Please use a valid corporate/institutional email and specify your intended use case.
The Transparent-460 database is available for non-commercial research purposes only.
-
All images of the Transparent-460 database are obtained from the Internet and are not the authors' property. The authors are not responsible for these images' content or meaning.
-
You agree not to reproduce, duplicate, copy, sell, trade, resell or exploit any portion of the images and any portion of derived data for any commercial purposes.
-
You agree not to further copy, publish or distribute any portion of the Transparent-460 database. Except for internal use at a single site within the same organization, it is allowed to make copies of the dataset.
-
The authors reserve the right to terminate your access to the Transparent-460 database at any time.
@inproceedings{cai2022TransMatting,
title={TransMatting: Enhancing Transparent Objects Matting with Transformers},
author={Cai, Huanqia and Xue, Fanglei, and Xu, Lele and Guo, Lili},
booktitle={European Conference on Computer Vision (ECCV)},
year={2022},
}