Research map / Video / Video-language understanding
Hybrid-Tower: Fine-grained Pseudo-query Interaction and Generation for Text-to-Video Retrieval
Bangxiang Lan, Ruobing Xie, Ruixiang Zhao, Xingwu Sun, Zhanhui Kang, Gang Yang, Xirong Li
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 reconstruction loss supervising pseudo-query with real text features (ablation) | PIG · MSRVTT-1k | R@1: 45.8 -> 48.6 | Improved |
| use video, frame and informative patch tokens as generator input instead of video token only (ablation) | PIG · MSRVTT-1k | R@1: 46.3 -> 48.6 | Improved |
| add informative patch tokens to video and frame tokens (ablation) | PIG · MSRVTT-1k | R@1: 47.5 -> 48.6 | Improved |
| add video-level token to frame and patch tokens (ablation) | PIG · MSRVTT-1k | R@1: 48.1 -> 48.6 | Improved |
| use causal attention transformer generator instead of MLP (ablation) | PIG · MSRVTT-1k | R@1: 44.1 -> 48.6 | Improved |
| use causal attention transformer generator instead of Q-former (ablation) | PIG · MSRVTT-1k | R@1: 47.6 -> 48.6 | Improved |
| use XPool attention fusioner instead of cross attention (ablation) | PIG · MSRVTT-1k | R@1: 42.2 -> 48.6 | Improved |
| use XPool attention fusioner instead of co-attention (ablation) | PIG · MSRVTT-1k | R@1: 44.5 -> 48.6 | Improved |
| use learned pseudo-query generator instead of ShareGPT4Video captions (ablation) | PIG · MSRVTT-1k | R@1: 43.8 -> 48.6 | Improved |
| use learned pseudo-query generator instead of m-PLUG2 captions (ablation) | PIG · MSRVTT-1k | R@1: 47.4 -> 48.6 | Improved |
| add PIG hybrid-tower pseudo-query interaction to CLIP-ViP backbone | CLIP-ViP B/32 · MSRVTT-1k | R@1: 46.0 -> 48.6 | Improved |
| add PIG to CLIP-ViP backbone | CLIP-ViP B/32 · MSVD | R@1: 45.1 -> 47.9 | Improved |
Closest papers by meaning
- Video-ColBERT: Contextualized Late Interaction for Text-to-Video Retrieval (CVPR 2025 · similarity 0.87)
- TempMe: Video Temporal Token Merging for Efficient Text-Video Retrieval (ICLR 2025 · similarity 0.87)
- Bridging the Semantic Granularity Gap Between Text and Frame Representations for Partially Relevant Video Retrieval (AAAI 2025 · similarity 0.86)
- Towards Efficient and Effective Text-to-Video Retrieval with Coarse-to-Fine Visual Representation Learning (AAAI 2024 · similarity 0.86)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 2025 · similarity 0.86)
- Text Proxy: Decomposing Retrieval from a 1-to-N Relationship into N 1-to-1 Relationships for Text-Video Retrieval (AAAI 2025 · similarity 0.86)
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.
