Research map / Video / Video segmentation and tracking
Motion-Focused Contrastive Learning of Video Representations
Rui Li, Yiheng Zhang, Zhaofan Qiu, Ting Yao, Dong Liu, Tao Mei
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 motion-focused temporal augmentation (TA) (ablation) | MCL baseline · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 77.98 -> 78.26 | Improved |
| add motion-focused temporal augmentation (TA) (ablation) | MCL baseline · UCF101 linear protocol (R2D) | top-1 accuracy (%): 76.51 -> 76.47 | Hurt |
| add motion-focused spatial augmentation (SA) (ablation) | MCL baseline · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 77.98 -> 79.54 | Improved |
| add both TA and SA on top of the baseline (ablation) | MCL baseline · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 77.98 -> 79.64 | Improved |
| add temporal motion alignment loss Lt (ablation) | MCL with TA+SA · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 79.64 -> 80.26 | Improved |
| add spatial motion alignment loss Ls (ablation) | MCL with TA+SA · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 79.64 -> 80.83 | Improved |
| add spatio-temporal motion alignment loss Lst (ablation) | MCL with TA+SA · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 79.64 -> 81.47 | Improved |
| fuse the three alignment losses Lt, Ls, Lst (ablation) | MCL with TA+SA · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 79.64 -> 81.91 | Improved |
| add spatial alignment loss Ls (ablation) | MCL with TA+SA · UCF101 linear protocol (R2D) | top-1 accuracy (%): 77.16 -> 77.84 | Improved |
| align weighted gradient map (MAL-v3) instead of feature map (MAL-v1) (ablation) | MCL · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 80.57 -> 81.91 | Improved |
| align weighted gradient map (MAL-v3) instead of weighted feature map (MAL-v2) (ablation) | MCL · UCF101 linear protocol (R(2+1)D) | top-1 accuracy (%): 80.99 -> 81.91 | Improved |
| use MCL instead of SeCo | MCL · Kinetics-400 linear protocol (R(2+1)D) | top-1 accuracy (%): 62.50 -> 66.62 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SeCo — SeCo: Exploring Sequence Supervision for Unsupervised Representation Learning (AAAI 2021) · better on 5 of 5 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
- Motion Sensitive Contrastive Learning for Self-Supervised Video Representation (ECCV 2022 · similarity 0.92)
- MaCLR: Motion-Aware Contrastive Learning of Representations for Videos (ECCV 2022 · similarity 0.90)
- Spatiotemporal Contrastive Video Representation Learning (CVPR 2021 · similarity 0.90)
- Self-Supervised Video Representation Learning in a Heuristic Decoupled Perspective (IJCV 2026 · similarity 0.89)
- Contrast and Order Representations for Video Self-Supervised Learning (ICCV 2021 · similarity 0.89)
- Composable Augmentation Encoding for Video Representation 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.
