Research map / Label-efficient and robust learning / Self-supervised and contrastive
Progressive Mask Distillation for Self-supervised Video Representation
Kewei Wu, Chong Liang, Zhao Xie, Dan Guo
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 progressive student distillation (increasing mask ratio across phases) (ablation) | MVD base · SSv2 (ViT-B) | Top-1 accuracy (%): 72.5 -> 73.0 | Improved |
| add difficulty-aware region enhancement (ablation) | MVD base · SSv2 (ViT-B) | Top-1 accuracy (%): 72.5 -> 72.8 | Improved |
| add cross-layer feature alignment (ablation) | MVD base · SSv2 (ViT-B) | Top-1 accuracy (%): 72.5 -> 72.7 | Improved |
| combine all three modules (ablation) | MVD base · SSv2 (ViT-B) | Top-1 accuracy (%): 72.5 -> 73.7 | Improved |
| use four progressive phases instead of a single 90% mask ratio (ablation) | PMD pre-training · SSv2 (ViT-B) | Top-1 accuracy (%): 73.1 -> 73.7 | Improved |
| smoother mask ratio progression (75->80->90 vs 75->85->90) (ablation) | PMD pre-training · SSv2 (ViT-B) | Top-1 accuracy (%): 73.5 -> 73.6 | Improved |
| use higher single mask ratio (75% to 90%) (ablation) | single-phase pre-training · SSv2 (ViT-B) | Top-1 accuracy (%): 72.7 -> 73.1 | Improved |
| add smoothed patch loss and region weights together (ablation) | difficulty-aware region enhancement · SSv2 (ViT-B) | Top-1 accuracy (%): 72.9 -> 73.7 | Improved |
| add both shallow->middle and middle->deep alignment (ablation) | cross-layer feature alignment · SSv2 (ViT-B) | Top-1 accuracy (%): 73.2 -> 73.7 | Improved |
| increase epochs per stage beyond 32 (ablation) | difficulty-aware region enhancement · SSv2 (ViT-B) | Top-1 accuracy (%): 73.7 -> 73.7 | No clear effect |
| increase pre-training epochs from 400 to 800 (ablation) | PMD · SSv2 (ViT-B) | Top-1 accuracy (%): 73.7 -> 73.7 | No clear effect |
| use image plus video teachers instead of one teacher (ablation) | PMD · SSv2 (ViT-S student) | Top-1 accuracy (%): 73.4 -> 73.7 | Improved |
Closest papers by meaning
- Masked Video Distillation: Rethinking Masked Feature Modeling for Self-Supervised Video Representation Learning (CVPR 2023 · similarity 0.92)
- Asymmetric Masked Distillation for Pre-Training Small Foundation Models (CVPR 2024 · similarity 0.90)
- Video Language Model Pretraining with Spatio-temporal Masking (CVPR 2025 · similarity 0.89)
- Exploring Target Representations for Masked Autoencoders (ICLR 2024 · similarity 0.89)
- Hybrid Distillation: Connecting Masked Autoencoders with Contrastive Learners (ICLR 2024 · similarity 0.88)
- From Prototypes to General Distributions: An Efficient Curriculum for Masked Image Modeling (CVPR 2025 · 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.
