Open JosephPai opened 1 year ago
Hi @JosephPai, thanks for the suggestion! I believe this work would be more suited for the awesome list https://github.com/Qinying-Liu/Awesome-Open-Vocabulary-Semantic-Segmentation. In ours we focus on fully-unsupervised methods which are mainly class-agnostic. Hope that makes sense, Best,
Hi @osimeoni , thanks for the suggestion! I think our work somehow lies in the middle ground between object localization and semantic segmentation. If you look into the the detail, you will see that the localization part is actually achieved in an unsupervised way without category. The labeling is an additional feature based on it. And it turns out that the semantic labeling can further improve the localization.
But anyway, as stated in the title, our main target would still be object localization. We are still some distance away from unsupervised semantic segmentation, which would be an exciting future work :) So I insist on recommending that you add this paper. As far as I know, there are not too many papers on object localization in videos.
Unsupervised Open-Vocabulary Object Localization in Videos from ICCV 2023
Thanks!