Research map / Video / Video segmentation and tracking
Learning From Untrimmed Videos: Self-Supervised Video Representation Learning With Hierarchical Consistency
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 TCL to VCL and use gradual sampling (ablation) | HiCo pretrained on UK400 · HMDB51 (LFT) | accuracy: 41.9 -> 47.5 | Helped |
| add VCL to standard contrastive learning (ablation) | HiCo pretrained on UK400 · HMDB51 (LFT) | accuracy: 34.7 -> 37.8 | Helped |
| add TCL to standard contrastive learning (ablation) | HiCo pretrained on UK400 · HMDB51 (LFT) | accuracy: 34.7 -> 34.8 | No clear effect |
| combine visual consistency and topical consistency learning (ablation) | HiCo pretrained on UK400 · HMDB51 (LFT) | accuracy: 37.8 -> 41.9 | Helped |
| add gradual sampling (ablation) | VCL + TCL pretrained on UK400 · UCF101 (LFT) | accuracy: 68.8 -> 70.7 | Helped |
| include the topical clip in the VCL negative sample pool (ablation) | HiCo pretrained on UK400 · HMDB51 (FT) | accuracy: 48.8 -> 50.5 | Helped |
| use bidirectional instead of unidirectional topical-feature concatenation (ablation) | HiCo pretrained on UK400 · HMDB51 (FT) | accuracy: 52.4 -> 54.1 | Helped |
| set the gradual-sampling upper bound Δ to 0.0 rather than 1.0 (ablation) | HiCo pretrained on UK400 · HMDB51 (FT) | accuracy: 54.1 -> 51.9 | Hurt |
| increase topical-pair distance from 0 to +∞ (ablation) | HiCo pretrained on UK400 · HMDB51 (FT) | accuracy: 51.5 -> 54.1 | Helped |
| pretrain with HiCo rather than standard contrastive learning | S3D-G pretrained on UK400 · ActivityNet action localization | AUC: 63.0% -> 67.1% | Helped |
| pretrain with HiCo on untrimmed rather than trimmed Kinetics-400 | S3D-G action recognition · HMDB51 (LFT) | accuracy: 35.5 -> 44.9 | Helped |
Closest papers by meaning
- Spatiotemporal Contrastive Video Representation Learning (CVPR 2021 · similarity 0.88)
- Scene Consistency Representation Learning for Video Scene Segmentation (CVPR 2022 · similarity 0.88)
- From Static to Dynamic: Exploring Self-supervised Image-to-Video Representation Transfer Learning (CVPR 2026 · similarity 0.88)
- Contextualized Spatio-Temporal Contrastive Learning With Self-Supervision (CVPR 2022 · similarity 0.87)
- Self-Supervised Spatio-Temporal Representation Learning for Videos by Predicting Motion and Appearance Statistics (CVPR 2019 · similarity 0.87)
- Learning by Aligning Videos in Time (CVPR 2021 · 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.