Research map / Video / Video segmentation and tracking
Self-Supervised Learning of Video-Induced Visual Invariances
Check 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 shot-level temporal exemplars to frame-based exemplar SSL (ablation) | Ex-YT-F on YouTube-8M · VTAB v1, 1000 examples | mean accuracy: 59.4 -> 61.3 | Helped |
| Add video-level prediction across 4 shots (ablation) | Ex-YT-S on YouTube-8M · VTAB v1, 1000 examples | mean accuracy: 61.3 -> 62.5 | Helped |
| Widen the VIVI-Ex(4) architecture by 3× (ablation) | VIVI-Ex(4) · VTAB v1, 1000 examples | mean accuracy: 62.5 -> 63.3 | Helped |
| Replace the usual augmentation policy with AutoAugment (ablation) | VIVI-Ex(4) on VTAB · VTAB v1, 1000 examples | mean accuracy: 62.5 -> 63.0 | Helped |
| Replace the usual augmentation policy with AutoAugment (ablation) | VIVI-Rot on VTAB · VTAB v1, 1000 examples | mean accuracy: 58.9 -> 59.9 | Helped |
| Use 4 shots rather than 2 for InfoNCE video prediction (ablation) | VIVI-Ex on VTAB · VTAB v1, 1000 examples | mean accuracy: 62.0 -> 62.5 | Helped |
| Use shot-order prediction rather than InfoNCE video prediction (ablation) | VIVI-Ex with 2 shots · VTAB v1, 1000 examples | mean accuracy: 62.0 -> 61.9 | No clear effect |
| Co-train with 10% labeled ImageNet data (ablation) | VIVI-Ex(4) · VTAB v1, 1000 examples | mean accuracy: 62.5 -> 67.2 | Helped |
| Co-train with the full labeled ImageNet training set (ablation) | VIVI-Ex(4) · VTAB v1, 1000 examples | mean accuracy: 62.5 -> 69.4 | Helped |
| Increase co-trained model capacity and use AutoAugment on ImageNet (ablation) | VIVI-Ex(4)-Co(100%) · VTAB v1, 1000 examples | mean accuracy: 69.4 -> 71.7 | Helped |
| Co-train VIVI with labeled ImageNet data (ablation) | VIVI-Ex(4) · VTAB v1, 1000 examples | accuracy | Mixed |
| Co-train with labeled ImageNet data (ablation) | VIVI-Ex(4) on ImageNet-Vid-Robust · ImageNet-Vid-Robust | Δ accuracy: 18.1 -> 15.9 | No clear effect |
Closest papers by meaning
- Demystifying Contrastive Self-Supervised Learning: Invariances, Augmentations and Dataset Biases (NeurIPS 2020 · similarity 0.89)
- How Well Do Self-Supervised Models Transfer? (CVPR 2021 · similarity 0.89)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.88)
- Self-Supervised Cross-View Correspondence with Predictive Cycle Consistency (CVPR 2025 · similarity 0.86)
- PooDLe🐩: Pooled and dense self-supervised learning from naturalistic videos (ICLR 2025 · similarity 0.86)
- In-Context Symmetries: Self-Supervised Learning through Contextual World Models (NeurIPS 2024 · similarity 0.86)
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.