Research map / Video / Video segmentation and tracking
Contrastive Masked Autoencoders for Self-Supervised Video Hashing
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 |
|---|---|---|---|
| combine masked-frame reconstruction and contrastive learning in one hashing framework | ConMH · FCVID, 16 bits | mAP@5: 0.287 -> 0.350; 0.262 -> 0.350 | Helped |
| add temporal masking to reconstruction-only training (ablation) | ConMH · FCVID, 16 bits | mAP@5: 0.242 -> 0.287 | Helped |
| use hashing-oriented objectives on hash codes instead of general feature learning (ablation) | ConMH · FCVID, 16 bits | mAP@5: 0.274 -> 0.350 | Helped |
| use a masking ratio in the 50%-to-75% range rather than a low or extremely high ratio (ablation) | ConMH · FCVID, 64 bits | mAP@K | Mixed |
| use entire-frame masking instead of tube-based or token-based masking (ablation) | ConMH · FCVID, 64 bits | mAP@5: 0.387 -> 0.524; 0.412 -> 0.524 | Helped |
| use long-and-short sampling rather than non-overlapped sampling (ablation) | ConMH · FCVID, 64 bits | mAP@5: 0.509 vs 0.524 | Hurt |
| use overlapped sampling rather than non-overlapped sampling (ablation) | ConMH · FCVID, 64 bits | mAP@5: 0.528 vs 0.524 | No clear effect |
| use a small decoder rather than a deeper or substantially wider decoder (ablation) | ConMH · FCVID, 64 bits | mAP@20: 0.373 (depth 2) vs 0.366 (depth 12); 0.373 (width 192) vs 0.370 (width 1024) | Mixed |
| use a larger encoder instead of ViT-small (ablation) | ConMH · FCVID, 64 bits | mAP@20: 0.373 -> 0.397; 0.373 -> 0.401 | Helped |
| replace CNN features with a Swin Transformer backbone | ConMH · FCVID, 64 bits | mAP@5: 0.524 -> 0.675 | Helped |
Closest papers by meaning
- Efficient Self-Supervised Video Hashing with Selective State Spaces (AAAI 2025 · similarity 0.93)
- AutoSSVH: Exploring Automated Frame Sampling for Efficient Self-Supervised Video Hashing (CVPR 2025 · similarity 0.92)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.86)
- An Asymmetric Augmented Self-Supervised Learning Method for Unsupervised Fine-Grained Image Hashing (CVPR 2024 · similarity 0.86)
- EVEREST: Efficient Masked Video Autoencoder by Removing Redundant Spatiotemporal Tokens (ICML 2024 · similarity 0.85)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.85)
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.