Research map / Video / Video-language understanding
HiT: Hierarchical Transformer With Momentum Contrast for Video-Text Retrieval
Song Liu, Haoqi Fan, Shengsheng Qian, Yiru Chen, Wenkui Ding, Zhongyuan Wang
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 momentum cross-modal contrast (4,096 queue) (ablation) | HiT · MSR-VTT | rsum: 306.5 -> 320.3 | Improved |
| increase memory bank from 4,096 to 8,192 (ablation) | HiT · MSR-VTT | rsum: 320.3 -> 316.2 | Hurt |
| add momentum key encoders instead of query encoders for bank (ablation) | HiT · MSR-VTT | text-to-video R@5: 48.4 -> 59.2 | Improved |
| add feature-level matching to semantic-level matching (ablation) | HiT · MSR-VTT | text-to-video R@1: 23.5 -> 27.0 | Improved |
| add semantic-level matching to feature-level matching (ablation) | HiT · MSR-VTT | text-to-video R@1: 25.1 -> 27.0 | Improved |
| add a third matching level (3-level-a) (ablation) | HiT · MSR-VTT | text-to-video R@1: 27.0 -> 28.5 | No clear effect |
| add four matching levels (ablation) | HiT · MSR-VTT | text-to-video R@1: 27.0 -> 27.1 | No clear effect |
| use InfoNCE instead of triplet ranking loss (ablation) | HiT · MSR-VTT | text-to-video R@1: 25.6 -> 27.7 | Improved |
| set temperature to 0.07 instead of 0.7 (ablation) | HiT · MSR-VTT | rsum: 155.4 -> 320.3 | Improved |
| set temperature to 0.07 instead of 0.0007 (ablation) | HiT · MSR-VTT | rsum: 285.1 -> 320.3 | Improved |
| add appearance expert to motion and audio (ablation) | HiT · MSR-VTT | text-to-video R@1: 28.1 -> 27.7 | No clear effect |
| add audio expert to motion (ablation) | HiT · MSR-VTT | text-to-video R@1: 25.1 -> 28.1 | Improved |
Closest papers by meaning
- CrossCLR: Cross-Modal Contrastive Learning for Multi-Modal Video Representations (ICCV 2021 · similarity 0.88)
- Multi-modal Transformer for Video Retrieval (ECCV 2020 · similarity 0.88)
- Learning Dynamic Similarity by Bidirectional Hierarchical Sliding Semantic Probe for Efficient Text Video Retrieval (AAAI 2025 · similarity 0.87)
- MV-Adapter: Multimodal Video Transfer Learning for Video Text Retrieval (CVPR 2024 · similarity 0.85)
- Diffusion-Inspired Truncated Sampler for Text-Video Retrieval (NeurIPS 2024 · similarity 0.85)
- Recurrence-Enhanced Vision-and-Language Transformers for Robust Multimodal Document Retrieval (CVPR 2025 · 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.
