Research map / Video / Video-language understanding
Lightweight Attentional Feature Fusion: A New Baseline for Text-to-Video Retrieval
Fan Hu, Aozhu Chen, Ziyue Wang, Fangming Zhou, Jianfeng Dong, 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 |
|---|---|---|---|
| replace MHSA with LAFF block (ablation) | text-video retrieval, 5 text features · MSR-VTT MV-test3k | mAP: 0.294 -> 0.358 | Improved |
| add the attentional layer to Attention-free fusion (LAFF) (ablation) | feature fusion · MSR-VTT MV-test3k | mAP: 0.326 -> 0.358 | Improved |
| replace concatenation with Attention-free fusion (ablation) | feature fusion · MSR-VTT MV-test3k | mAP: 0.247 -> 0.326 | Improved |
| add multi-loss to MHSA (ablation) | MHSA fusion · MSR-VTT MV-test3k | mAP: 0.294 -> 0.306 | Improved |
| add BERT text feature to concatenation (W2VV++) (ablation) | concatenation fusion · MSR-VTT MV-test3k | mAP: 0.310 -> 0.247 | Hurt |
| add BERT text feature to LAFF (ablation) | LAFF · MSR-VTT MV-test3k | mAP: 0.358 -> 0.358 | No clear effect |
| add CLIP text feature to LAFF (ablation) | LAFF · MSR-VTT MV-test3k | mAP: 0.276 -> 0.358 | Improved |
| use combined per-space loss instead of single loss (ablation) | LAFF · MSR-VTT MV-test3k | mAP: 0.324 -> 0.358 | Improved |
| increase number of common spaces h from 1 to 8 (ablation) | LAFF · MSR-VTT MV-test3k | mAP: 0.352 -> 0.358 | Improved |
| increase h from 8 to 16 (ablation) | LAFF · MSR-VTT MV-test3k | mAP: 0.358 -> 0.356 | No clear effect |
| keep only top-3 video and text features chosen by LAFF weights (ablation) | LAFF · MSR-VTT MV-test3k | mAP: 0.358 -> 0.353 | No clear effect |
| use multi-level LAFF-ml instead of LAFF | LAFF · MSR-VTT MV-test3k | R@1: 28.0 -> 29.1 | Improved |
Closest papers by meaning
- Towards Efficient and Effective Text-to-Video Retrieval with Coarse-to-Fine Visual Representation Learning (AAAI 2024 · similarity 0.87)
- Bridging the Semantic Granularity Gap Between Text and Frame Representations for Partially Relevant Video Retrieval (AAAI 2025 · similarity 0.85)
- TempMe: Video Temporal Token Merging for Efficient Text-Video Retrieval (ICLR 2025 · similarity 0.85)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 2025 · similarity 0.85)
- Learning Dynamic Similarity by Bidirectional Hierarchical Sliding Semantic Probe for Efficient Text Video Retrieval (AAAI 2025 · similarity 0.84)
- DGL: Dynamic Global-Local Prompt Tuning for Text-Video Retrieval (AAAI 2024 · similarity 0.84)
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.
