Research map / Video / Video segmentation and tracking
Masked Motion Encoding for Self-Supervised Video Representation Learning
Read the paperCheck 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 |
|---|---|---|---|
| Replace appearance reconstruction with motion-trajectory reconstruction | VideoMAE-style masked video modeling · Something-Something V2, K400 pre-training | Acc.@1: 68.2% to 70.5% | Helped |
| Use motion-trajectory reconstruction | VideoMAE-style masked video modeling · Kinetics-400 | Acc.@1: +0.5% | Helped |
| Add position features from long-term trajectories to HOG reconstruction (ablation) | HOG reconstruction · Something-Something V2 ablation | Acc.@1: 61.1 -> 63.5 | Helped |
| Add trajectory-aligned shape features (ablation) | Trajectory position features · Something-Something V2 ablation | Acc.@1: 63.5 -> 64.1 | Helped |
| Add dense optical-flow prediction to HOG reconstruction (ablation) | HOG reconstruction · Something-Something V2 ablation | Acc.@1: 61.1 -> 60.9 | Hurt |
| Add HOF reconstruction to HOG (ablation) | HOG reconstruction · Something-Something V2 ablation | Acc.@1: 61.1 -> 61.3 | Helped |
| Add MBH reconstruction to HOG (ablation) | HOG reconstruction · Something-Something V2 ablation | Acc.@1: 61.1 -> 61.4 | Helped |
| Interpolate dense motion trajectories (ablation) | MME · Something-Something V2 ablation | Acc.@1: 62.2 -> 64.1 | Helped |
| Interpolate HOG targets (ablation) | HOG reconstruction · Something-Something V2 ablation | Acc.@1: 61.1 -> 60.9 | Hurt |
| Interpolate HOG and MBH targets (ablation) | HOG + MBH reconstruction · Something-Something V2 ablation | Acc.@1: 61.3 -> 61.4 | No clear effect |
| Use trajectory length L=6 instead of L=4 or L=8 (ablation) | MME · Something-Something V2 ablation | Acc.@1: 63.2 (L=4); 64.1 (L=6); 63.3 (L=8) | Helped |
| Add per-patch trajectory normalization (ablation) | MME · Something-Something V2 ablation | Acc.@1: 62.1 -> 64.1 | Helped |
Closest papers by meaning
- Video Language Model Pretraining with Spatio-temporal Masking (CVPR 2025 · similarity 0.88)
- SMILE: Infusing Spatial and Motion Semantics in Masked Video Learning (CVPR 2025 · similarity 0.88)
- Through-The-Mask: Mask-based Motion Trajectories for Image-to-Video Generation (CVPR 2025 · similarity 0.87)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.87)
- Rethinking Masked Data Reconstruction Pretraining for Strong 3D Action Representation Learning (AAAI 2025 · similarity 0.87)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · 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.