Research map / Video / Video-language understanding
CoVR-2: Automatic Data Construction for Composed Video Retrieval
Lucas Ventura, Antoine Yang, Cordelia Schmid, Gül Varol
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 caption retrieval loss (ablation) | CoVR-BLIP-2 · CIRR (trained on WebVid-CoVR) | R@1: 34.43 -> 41.42 | Improved |
| add caption retrieval loss (ablation) | CoVR-BLIP-2 · FashionIQ | R@10: 27.78 -> 36.81 | Improved |
| add caption retrieval loss (ablation) | CoVR-BLIP-2 · CIRCO | mAP@5: 25.06 -> 27.84 | Improved |
| add caption retrieval loss (ablation) | CoVR-BLIP-2 · WebVid-CoVR-Test | R@1: 60.09 -> 59.82 | No clear effect |
| use BLIP-2 cross-attention fusion instead of MLP fusion (ablation) | CoVR-BLIP-2 · WebVid-CoVR-Test | R@1: 51.88 -> 59.82 | Improved |
| finetune on WebVid-CoVR instead of frozen embeddings (ablation) | CoVR-BLIP-2 · WebVid-CoVR-Test | R@1: 45.66 -> 59.82 | Improved |
| combine visual and text instead of visual only (ablation) | CoVR-BLIP-2 · WebVid-CoVR-Test | R@1: 36.03 -> 59.82 | Improved |
| use finetuned MTG-LLM instead of rule-based modification text (ablation) | CoVR-BLIP-2 · WebVid-CoVR-Test | R@1: 39.08 -> 59.82 | Improved |
| use finetuned MTG-LLM instead of GPT-paraphrased rule text (ablation) | CoVR-BLIP-2 · CIRR | R@1: 32.89 -> 41.42 | Improved |
| iterate over target videos instead of triplets (ablation) | CoVR-BLIP-2 · WebVid-CoVR-Test | R@1: 53.79 -> 55.24 | Improved |
| add HN-NCE hard negative up-sampling (ablation) | CoVR-BLIP-2 · WebVid-CoVR-Test | R@1: 55.24 -> 59.82 | Improved |
| add CC-CoIR image data to WebVid-CoVR pretraining (ablation) | CoVR-BLIP-2 · CIRR zero-shot | R@1: 41.42 -> 43.74 | Improved |
Closest papers by meaning
- CoVR: Learning Composed Video Retrieval from Web Video Captions (AAAI 2024 · similarity 0.98)
- Composed Video Retrieval via Enriched Context and Discriminative Embeddings (CVPR 2024 · similarity 0.93)
- From Play to Replay: Composed Video Retrieval for Temporally Fine-Grained Videos (NeurIPS 2025 · similarity 0.90)
- Data Roaming and Quality Assessment for Composed Image Retrieval (AAAI 2024 · similarity 0.90)
- CoLLM: A Large Language Model for Composed Image Retrieval (CVPR 2025 · similarity 0.88)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.88)
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.
