Research map / Video / Action recognition
Self-supervised Video Representation Learning by Pace Prediction
Jiangliu Wang, Jianbo Jiao, Yun-Hui Liu
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 color jittering to pace prediction (ablation) | R(2+1)D pace prediction · UCF101 | accuracy: 72.0 -> 73.9 | Improved |
| raise max pace from 4 to 6 classes (ablation) | pace prediction p=[1,x] · UCF101 | accuracy: 72.0 -> 71.1 | Hurt |
| pace prediction pre-training vs random init (ablation) | R(2+1)D · UCF101 | accuracy: 56.0 -> 72.0 | Improved |
| add contrastive learning with same context to pace prediction (ablation) | R(2+1)D · UCF101 | accuracy: 73.9 -> 75.8 | Improved |
| add contrastive learning with same pace to pace prediction (ablation) | R(2+1)D · UCF101 | accuracy: 73.9 -> 73.6 | Hurt |
| add nonlinear fc layer before contrastive space (ablation) | R(2+1)D pace + same-context contrastive · HMDB51 | accuracy: 35.0 -> 35.9 | Improved |
| use same context instead of same pace for contrastive learning alone (ablation) | R(2+1)D · UCF101 | accuracy: 59.4 -> 67.3 | Improved |
| use slow pace instead of fast pace (ablation) | pace prediction · UCF101 | accuracy: 73.9 -> 72.6 | Hurt |
| use slow-fast absolute pace instead of fast pace (ablation) | pace prediction · UCF101 | accuracy: 73.9 -> 73.9 | No clear effect |
| increase pace step from 1 to 2 (ablation) | pace prediction · UCF101 | accuracy: 73.9 -> 74.9 | Improved |
| increase pace step from 2 to 3 (ablation) | pace prediction · UCF101 | accuracy: 74.9 -> 74.7 | Hurt |
| add random backwards play (4 classes) (ablation) | pace prediction · UCF101 | accuracy: 73.9 -> 73.0 | Hurt |
Closest papers by meaning
- Self-Supervised Video Representation Learning by Context and Motion Decoupling (CVPR 2021 · similarity 0.90)
- Self-Supervised Video Representation Learning via Latent Time Navigation (AAAI 2023 · similarity 0.89)
- SpeedNet: Learning the Speediness in Videos (CVPR 2020 · similarity 0.89)
- Learning Video Representations by Transforming Time (ECCV 2020 · similarity 0.89)
- Self-Supervised Spatio-Temporal Representation Learning for Videos by Predicting Motion and Appearance Statistics (CVPR 2019 · similarity 0.89)
- Contrast and Order Representations for Video Self-Supervised Learning (ICCV 2021 · 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.
