Research map / Video / Video-language understanding
Bidirectional Likelihood Estimation with Multi-Modal Large Language Models for Text-Video Retrieval
Dohwan Ko, Ji Soo Lee, Minhyuk Choi, Zihang Meng, Hyunwoo J. Kim
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 query likelihood estimation to candidate likelihood (QLE vs CLE) (ablation) | BLiM no CPN · avg R@1 four text-video datasets | R@1: 27.3 -> 60.5 | Improved |
| combine candidate and query likelihoods (BLE vs QLE) (ablation) | BLiM no CPN · avg R@1 four text-video datasets | R@1: 60.5 -> 61.3 | Improved |
| add Candidate Prior Normalization to candidate likelihood (ablation) | CLE · avg R@1 four text-video datasets | R@1: 34.4 -> 59.2 (DiDeMo) | Improved |
| add Candidate Prior Normalization to bidirectional estimation (ablation) | BLiM BLE · avg R@1 four text-video datasets | R@1: 74.1 -> 81.3 (DiDeMo) | Improved |
| fine-tune with bidirectional likelihood instead of zero-shot BLiM- | BLiM · DiDeMo text-to-video | R@1: 69.8 -> 86.4 | Improved |
| rerank with BLiM instead of InternVideo2 6B | BLiM · DiDeMo text-to-video | R@1: 74.2 -> 86.4 | Improved |
| use BLiM instead of other MLLM rerankers (RagVL) | BLiM · DiDeMo text-to-video | R@1: 83.2 -> 86.4 | Improved |
| apply CPN decoding to VideoChat2 | VideoChat2 · VideoMME w/o subtitle | accuracy: 42.2 -> 47.1 | Improved |
| apply CPN decoding to LLaVA-Onevision | LLaVA-Onevision · VideoMME w/o subtitle | accuracy: 58.5 -> 61.7 | Improved |
| apply CPN decoding to InternVL2 | InternVL2 · MLVU | accuracy: 50.8 -> 55.1 | Improved |
| apply BLiM to text-image retrieval | BLiM · COCO image-to-text | R@1: 85.4 -> 84.2 | Hurt |
| apply CPN in text-to-video retrieval with CLE (ablation) | CLE · DiDeMo text-to-video | R@1: 45.1 -> 45.1 | No clear effect |
Closest papers by meaning
- Bridging Information Asymmetry in Text-video Retrieval: A Data-centric Approach (ICLR 2025 · similarity 0.88)
- LamRA: Large Multimodal Model as Your Advanced Retrieval Assistant (CVPR 2025 · similarity 0.88)
- MM-EMBED: UNIVERSAL MULTIMODAL RETRIEVAL WITH MULTIMODAL LLMS (ICLR 2025 · similarity 0.87)
- Text Proxy: Decomposing Retrieval from a 1-to-N Relationship into N 1-to-1 Relationships for Text-Video Retrieval (AAAI 2025 · similarity 0.87)
- BayesRAG: Probabilistic Mutual Evidence Corroboration for Multimodal Retrieval-Augmented Generation (NeurIPS 2026 · similarity 0.87)
- Video-RAG: Visually-aligned Retrieval-Augmented Long Video Comprehension (NeurIPS 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.
