English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In this paper, we propose a simple yet effective approach for self-supervisedvideo object segmentation (VOS). Our key insight is that the inherentstructural dependencies present in DINO-pretrained Transformers can beleveraged to establish robust spatio-temporal correspondences in videos.Furthermore, simple clustering on this correspondence cue is sufficient toyield competitive segmentation results. Previous self-supervised VOS techniquesmajorly resort to auxiliary modalities or utilize iterative slot attention toassist in object discovery, which restricts their general applicability andimposes higher computational requirements. To deal with these challenges, wedevelop a simplified architecture that capitalizes on the emerging objectnessfrom DINO-pretrained Transformers, bypassing the need for additional modalitiesor slot attention. Specifically, we first introduce a single spatio-temporalTransformer block to process the frame-wise DINO features and establishspatio-temporal dependencies in the form of self-attention. Subsequently,utilizing these attention maps, we implement hierarchical clustering togenerate object segmentation masks. To train the spatio-temporal block in afully self-supervised manner, we employ semantic and dynamic motion consistencycoupled with entropy normalization. Our method demonstrates state-of-the-artperformance across multiple unsupervised VOS benchmarks and particularly excelsin complex real-world multi-object video segmentation tasks such asDAVIS-17-Unsupervised and YouTube-VIS-19. The code and model checkpoints willbe released at https://github.com/shvdiwnkozbw/SSL-UVOS.

Betrayed by Attention: A Simple yet Effective Approach for Self-supervised Video Object Segmentation