Research map / Video / Video segmentation and tracking
Self-Supervised Video Hashing via Bidirectional Transformers
Check 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 |
|---|---|---|---|
| Use bidirectional transformers instead of CNN, LSTM, or BiLSTM sequence models | BTH · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.242 -> 0.248; 0.239 -> 0.248; 0.245 -> 0.248 | Helped |
| Add whole-sequence visual reconstruction (ablation) | BTH · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.218 -> 0.248 | Helped |
| Reconstruct the whole frame sequence rather than only masked visual words (ablation) | BTH visual cloze task · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.228 -> 0.248 | Helped |
| Mask some input frame features during training (ablation) | BTH visual cloze task · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.240 -> 0.248 | Helped |
| Add pairwise similarity reconstruction (ablation) | BTH · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.229 -> 0.248 | Helped |
| Construct the proposed pairwise similarity connections instead of using the NPH similarity matrix (ablation) | BTH similarity reconstruction · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.238 -> 0.248 | Helped |
| Add cluster alignment using pseudo centers (ablation) | BTH · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.218 -> 0.248 | Helped |
| Construct pseudo centers from the training data rather than randomly (ablation) | BTH cluster alignment · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.230 -> 0.248 | Helped |
| Use a 128-D rather than 64-D bottleneck to construct pseudo centers (ablation) | BTH cluster alignment · FCVID, 32-bit codes, mAP@20 | mAP@20: 0.238 -> 0.248 | Helped |
| Update the pseudo center set during training (ablation) | BTH · FCVID, 64-bit codes, mAP@5 and mAP@60 | mAP@5 and mAP@60 | Mixed |
| Train on FCVID and test on YFCC rather than train and test on YFCC | BTH · YFCC, 64-bit codes, mAP@20 | mAP@20: ↓5.7 | Hurt |
| Train on FCVID and test on YFCC rather than train and test on YFCC | SSTH · YFCC, 64-bit codes, mAP@20 | mAP@20: ↓6.3 | Hurt |
Closest papers by meaning
- Contrastive Masked Autoencoders for Self-Supervised Video Hashing (AAAI 2023 · similarity 0.91)
- Efficient Self-Supervised Video Hashing with Selective State Spaces (AAAI 2025 · similarity 0.90)
- AutoSSVH: Exploring Automated Frame Sampling for Efficient Self-Supervised Video Hashing (CVPR 2025 · similarity 0.88)
- Deep Supervised Discrete Hashing (NeurIPS 2017 · similarity 0.85)
- An Asymmetric Augmented Self-Supervised Learning Method for Unsupervised Fine-Grained Image Hashing (CVPR 2024 · similarity 0.84)
- Vision-guided Text Mining for Unsupervised Cross-modal Hashing with Community Similarity Quantization (AAAI 2025 · similarity 0.82)
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.