Research map / Video / Action recognition
Learning Video Representations by Transforming Time
Simon Jenni, Givi Meishvili, Paolo Favaro
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 speed-type classification loss (ablation) | Speed+Random+Periodic+Warp · UCF101 (conv frozen, UCF101 pre-training) | accuracy: 54.2 -> 60.6 | Improved |
| add speed-type classification loss (ablation) | Speed+Random · UCF101 (conv frozen, UCF101 pre-training) | accuracy: 44.5 -> 55.1 | Improved |
| add speed-type classification loss (ablation) | Speed+Periodic · UCF101 (conv frozen, UCF101 pre-training) | accuracy: 40.6 -> 56.5 | Improved |
| add speed-type classification loss (ablation) | Speed+Random+Warp · HMDB51 (fine-tuned, UCF101 pre-training) | accuracy: 36.6 -> 39.2 | Improved |
| add Warp transformation (ablation) | Speed+Random with speed loss · UCF101 (conv frozen, UCF101 pre-training) | accuracy: 55.1 -> 60.4 | Improved |
| add Random transformation (ablation) | Speed+Periodic+Warp with speed loss · UCF101 (conv frozen, UCF101 pre-training) | accuracy: 59.5 -> 60.6 | Improved |
| add Random transformation (ablation) | Speed+Periodic+Warp with speed loss · HMDB51 (fine-tuned, UCF101 pre-training) | accuracy: 39.0 -> 38.0 | Hurt |
| add Periodic transformation (ablation) | Speed+Random+Warp with speed loss · HMDB51 (fine-tuned, UCF101 pre-training) | accuracy: 39.2 -> 38.0 | Hurt |
| use all four transformations instead of speed only (ablation) | speed-type classification · UCF101 (conv frozen, UCF101 pre-training) | accuracy: 49.3 -> 60.6 | Improved |
| pre-train on Kinetics instead of UCF101 (ablation) | 3D-ResNet18 transfer · UCF101 | accuracy: 77.3 -> 79.3 | Improved |
| pre-train on Kinetics instead of UCF101 (ablation) | C3D transfer · UCF101 | accuracy: 68.3 -> 69.9 | Improved |
| replace supervised pre-training with temporal transformation pre-training | C3D · HMDB51 (fine-tuned, UCF101 pre-training) | accuracy: 28.8 -> 38.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SpeedNet — SpeedNet: Learning the Speediness in Videos (CVPR 2020) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Time Is MattEr: Temporal Self-supervision for Video Transformers (ICML 2022 · similarity 0.90)
- Unsupervised Learning of Long-Term Motion Dynamics for Videos (CVPR 2017 · similarity 0.89)
- Self-Supervised Video Representation Learning by Context and Motion Decoupling (CVPR 2021 · similarity 0.89)
- TimeBalance: Temporally-Invariant and Temporally-Distinctive Video Representations for Semi-Supervised Action Recognition (CVPR 2023 · similarity 0.89)
- Learning by Aligning Videos in Time (CVPR 2021 · similarity 0.89)
- Self-supervised Video Representation Learning by Pace Prediction (ECCV 2020 · similarity 0.89)
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.
