Research map / Video / Video segmentation and tracking
Spatial-Then-Temporal Self-Supervised Learning for Video Correspondence
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 pyramid-level frame reconstruction (ablation) | Temporal feature learning · DAVIS-2017 validation | J&F_m: 64.9 -> 65.6 | Helped |
| Add local correlation distillation from the finer pyramid level (ablation) | Temporal feature learning · DAVIS-2017 validation | J&F_m: 65.6 -> 68.4 | Helped |
| Add entropy-based selection to local correlation distillation (ablation) | Temporal feature learning · DAVIS-2017 validation | J&F_m: 68.4 -> 69.3 | Helped |
| Train spatially first, then temporally, rather than jointly (ablation) | ResNet-50 spatial-temporal learning · DAVIS-2017 validation | J&F_m: 68.6 -> 69.4 | Helped |
| Add global correlation distillation to retain spatial cues (ablation) | Spatial-then-Temporal · DAVIS-2017 validation | J&F_m: 69.4 -> 71.3 | Helped |
| Use EWC instead of global correlation distillation (ablation) | Spatial-then-Temporal · DAVIS-2017 validation | J&F_m: 69.6 vs 71.3 | Hurt |
| Train temporally first, then add spatial contrastive learning (ablation) | ResNet-50 temporal-spatial learning · DAVIS-2017 validation | J&F_m: 67.5 -> 66.9 | Hurt |
| Add EWC to temporal-then-spatial training (ablation) | Temporal-then-Spatial · DAVIS-2017 validation | J&F_m: 66.9 -> 67.0 | No clear effect |
| Use stride 4 rather than stride 8 for temporal feature learning | Ours, ResNet-18 · DAVIS-2017 validation | J&F_m: 70.7 -> 73.6 | Helped |
| Combine spatial contrastive pretraining with temporal feature learning | Ours, ResNet-18 · DAVIS-2017 validation | J&F_m: 69.3 -> 70.7 | Helped |
Closest papers by meaning
- Self-Supervised Spatio-Temporal Representation Learning for Videos by Predicting Motion and Appearance Statistics (CVPR 2019 · similarity 0.90)
- Joint-task Self-supervised Learning for Temporal Correspondence (NeurIPS 2019 · similarity 0.90)
- Spatiotemporal Contrastive Video Representation Learning (CVPR 2021 · similarity 0.89)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.89)
- Exposing the Self-Supervised Space-Time Correspondence Learning via Graph Kernels (AAAI 2023 · similarity 0.89)
- When the Future Becomes the Past: Taming Temporal Correspondence for Self-supervised Video Representation Learning (CVPR 2025 · similarity 0.88)
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.