Research map / Video / Video segmentation and tracking
Masked Video Distillation: Rethinking Masked Feature Modeling 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 pixel targets with high-level features from pretrained teachers | Masked video pretraining · Kinetics-400 | top-1 accuracy: 81.5 -> 82.7 | Helped |
| Replace pixel targets with high-level features from pretrained teachers | Masked video pretraining · Something-Something V2 | top-1 accuracy: 69.7 -> 72.5 | Helped |
| Use an image teacher rather than a video teacher (ablation) | MVD · Kinetics-400 | top-1 accuracy: 80.1 -> 80.4 | Helped |
| Use a video teacher rather than an image teacher (ablation) | MVD · Something-Something V2 | top-1 accuracy: 69.4 -> 70.0 | Helped |
| Add the complementary teacher type to use image and video co-teaching (ablation) | MVD with a single teacher · Kinetics-400 | top-1 accuracy: 80.4 -> 80.6 | Helped |
| Add the complementary teacher type to use image and video co-teaching (ablation) | MVD with a single teacher · Something-Something V2 | top-1 accuracy: 70.0 -> 70.7 | Helped |
| Add a pixel-reconstruction decoder branch (ablation) | MVD with image teacher · Something-Something V2 | top-1 accuracy: 68.7 -> 67.9 | Hurt |
| Add a pixel-reconstruction decoder branch (ablation) | MVD co-teaching · Something-Something V2 | top-1 accuracy: 70.1 -> 69.0 | Hurt |
| Replace the fixed pretrained teacher with a momentum encoder teacher (ablation) | MVD · Kinetics-400 | top-1 accuracy: 82.3 -> 80.5 | Hurt |
| Replace the fixed pretrained teacher with a momentum encoder teacher (ablation) | MVD · Something-Something V2 | top-1 accuracy: 71.4 -> 70.4 | Hurt |
| Initialize the momentum-encoder teacher method from IN-1K (ablation) | MVD with momentum encoder teacher · Kinetics-400 | top-1 accuracy: 80.5 -> 81.8 | Helped |
| Initialize the momentum-encoder teacher method from IN-1K (ablation) | MVD with momentum encoder teacher · Something-Something V2 | top-1 accuracy: 70.4 -> 70.8 | Helped |
Closest papers by meaning
- Asymmetric Masked Distillation for Pre-Training Small Foundation Models (CVPR 2024 · similarity 0.89)
- Exploring Target Representations for Masked Autoencoders (ICLR 2024 · similarity 0.89)
- Video Language Model Pretraining with Spatio-temporal Masking (CVPR 2025 · similarity 0.89)
- From Prototypes to General Distributions: An Efficient Curriculum for Masked Image Modeling (CVPR 2025 · similarity 0.88)
- VideoMAC: Video Masked Autoencoders Meet ConvNets (CVPR 2024 · similarity 0.88)
- Hybrid Distillation: Connecting Masked Autoencoders with Contrastive Learners (ICLR 2024 · 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.