Research map / Video / Video segmentation and tracking
Semantics Meets Temporal Correspondence: Self-supervised Object-centric Learning in Videos
Rui Qian, Shuangrui Ding, Xian Liu, Dahua Lin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add masked instance-identification stage after the semantic stage (ablation) | SMTC · DAVIS-2017-Unsupervised | J&F: 24.5 -> 40.5 | Improved |
| add semantic decomposition stage before instance identification (ablation) | SMTC · DAVIS-2016 | IoU: 52.1 -> 71.8 | Improved |
| increase number of Gaussian distributions N from 1 to 16 (ablation) | SMTC · DAVIS-2016 | IoU: 47.4 -> 71.8 | Improved |
| increase N from 16 to 32 (ablation) | SMTC · DAVIS-2016 | IoU: 71.8 -> 71.9 | No clear effect |
| fuse correlation map with RGB features instead of RGB only (ablation) | SMTC · DAVIS-2017-Unsupervised | J&F: 20.1 -> 40.5 | Improved |
| add semantic stage to correlation-only input (ablation) | SMTC · DAVIS-2016 | IoU: 59.4 -> 38.8 | Hurt |
| add semantic mask regularization L_reg (ablation) | SMTC · DAVIS-2016 | IoU: 61.3 -> 66.4 | Improved |
| add instance consistency loss L_obj (ablation) | SMTC · DAVIS-2017-Unsupervised | J&F: 24.5 -> 40.5 | Improved |
| increase sampling frame rate from 2 to 4 FPS (ablation) | SMTC · DAVIS-2016 | IoU: 66.3 -> 71.8 | Improved |
| increase frames per clip from 4 to 8 (ablation) | SMTC · DAVIS-2016 | IoU: 71.8 -> 71.9 | No clear effect |
| query slot initialization instead of random sampling on correlation map | slot attention on DINO features · DAVIS-2016 | IoU: 56.3 -> 21.5 | Hurt |
| random sampling instead of query slot initialization on RGB features | slot attention on DINO features · DAVIS-2016 | IoU: 58.1 -> 38.4 | Hurt |
Closest papers by meaning
- Rethinking Temporal Consistency in Video Object-Centric Learning: From Prediction to Correspondence (ICML 2026 · similarity 0.91)
- Betrayed by Attention: A Simple yet Effective Approach for Self-supervised Video Object Segmentation (ECCV 2024 · similarity 0.89)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.88)
- Temporally Consistent Object-Centric Learning by Contrasting Slots (CVPR 2025 · similarity 0.88)
- Adaptive Slot Attention: Object Discovery with Dynamic Slot Number (CVPR 2024 · similarity 0.88)
- Grounded Object-Centric Learning (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
