Research map / Video / Video-language understanding
Fine-Grained Text-to-Video Retrieval for Camera-Trap Data
Valentin Gabeff, Baptiste Maquignaz, Jiaxian Shan, Sepideh Mamooler, Gencer Sumbul, Blair Costelloe, Devis TUIA, Alexander Mathis
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 custom parsing library to the LLM code agent (ablation) | Qwen3-8B agent · Prompting-MammAlps (oracle) | F1: 0.44 -> 0.78 | Improved |
| add three in-context reasoning examples (ablation) | Qwen3-8B agent · Prompting-MammAlps (oracle) | F1: 0.78 -> 0.84 | Improved |
| add explicit label-space constraints with error feedback (ablation) | Qwen3-8B agent · Prompting-MammAlps (oracle) | F1: 0.84 -> 0.87 | Improved |
| use Qwen3-8B as code agent instead of other 7-8B LLMs (ablation) | Agent + Oracle · Prompting-MammAlps | F1: 0.42 -> 0.87 | Improved |
| use structured-text LLM agent retrieval instead of zero-shot VLMs | Agent + SALMA · Prompting-MammAlps (113 queries) | F1: 0.18 -> 0.37 | Improved |
| fine-tune CLIP4Clip on MammAlps-S2 query-video pairs | CLIP4Clip · Prompting-MammAlps (113 queries) | F1: 0.17 -> 0.26 | Improved |
| replace SALMA predictions with ground-truth annotations (oracle) | Agent · Prompting-MammAlps | F1: 0.34 -> 0.87 | Improved |
| remove InfoNCE contrastive loss on object queries (ablation) | SALMA-MOT · MammAlps-S2 tracking | IDF1: 84.5 -> 75.4 | Hurt |
| add second curriculum step (attribute heads) to SALMA-MOT | SALMA · MammAlps-S2 tracking | HOTA: 72.9 -> 69.0 | Hurt |
| use SALMA-MOT instead of MegaDetector+ByteTrack | SALMA-MOT · MammAlps-S2 tracking | HOTA / IDF1: HOTA 76.2 -> 72.9; IDF1 81.4 -> 84.5 | Mixed |
Closest papers by meaning
- Text Proxy: Decomposing Retrieval from a 1-to-N Relationship into N 1-to-1 Relationships for Text-Video Retrieval (AAAI 2025 · similarity 0.89)
- Bridging Information Asymmetry in Text-video Retrieval: A Data-centric Approach (ICLR 2025 · similarity 0.87)
- DGL: Dynamic Global-Local Prompt Tuning for Text-Video Retrieval (AAAI 2024 · similarity 0.87)
- TempMe: Video Temporal Token Merging for Efficient Text-Video Retrieval (ICLR 2025 · similarity 0.87)
- MUSE: Mamba Is Efficient Multi-scale Learner for Text-video Retrieval (AAAI 2025 · similarity 0.87)
- Needle In A Video Haystack: A Scalable Synthetic Evaluator for Video MLLMs (ICLR 2025 · similarity 0.87)
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.
