Research map / Video / Video-language understanding
CroPS: Improving Dense Retrieval with Cross-Perspective Positive Samples in Short-Video Search
Ao Xie, Jiahui Chen, Quanzhi Zhu, Xiaoze Jiang, Zhiheng Qin, Enyun Yu, Han 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 |
|---|---|---|---|
| add query-level positive augmentation (ablation) | dense retriever baseline · CPSQA CT | Recall@100: 59.6 -> 63.3 | Improved |
| add query-level positive augmentation (ablation) | dense retriever baseline · CPSQA QR | Recall@100: 33.0 -> 36.1 | Improved |
| add system-level positive augmentation (ablation) | CroPS with query-level aug. · CPSQA CT | Recall@100: 63.3 -> 65.7 | Improved |
| add system-level positive augmentation (ablation) | CroPS with query-level aug. · CPSQA QR | Recall@100: 36.1 -> 37.4 | Improved |
| add world-knowledge LLM positives (ablation) | CroPS with query and system aug. · CPSQA CT | Recall@100: 65.7 -> 69.1 | Improved |
| add world-knowledge LLM positives (ablation) | CroPS with query and system aug. · CPSQA QR | Recall@100: 37.4 -> 40.1 | Improved |
| use hierarchical labels with H-InfoNCE instead of binary labels (ablation) | CroPS · CPSQA CT | Recall@100: 59.9 -> 69.1 | Improved |
| use hierarchical labels with H-InfoNCE instead of binary labels (ablation) | CroPS · CPSQA QR | Recall@100: 32.1 -> 40.1 | Improved |
| give query-level positives the highest label 5 instead of 4 (ablation) | CroPS · CPSQA CT | Recall@100: 67.1 -> 69.1 | Improved |
| give query-level positives the highest label 5 instead of 4 (ablation) | CroPS · CPSQA QR | Recall@100: 38.4 -> 40.1 | Improved |
| use H-InfoNCE instead of InfoNCE (ablation) | CroPS · CPSQA training | training time: 178h -> 88h | Improved |
| use H-InfoNCE instead of Softmax-CE (ablation) | CroPS · CPSQA CT | Recall@100: 65.3 -> 69.1 | Improved |
Closest papers by meaning
- Retrieval-Augmented Perception: High-resolution Image Perception Meets Visual RAG (ICML 2025 · similarity 0.84)
- AdaVideoRAG: Omni-Contextual Adaptive Retrieval-Augmented Efficient Long Video Understanding (NeurIPS 2025 · similarity 0.84)
- Text Proxy: Decomposing Retrieval from a 1-to-N Relationship into N 1-to-1 Relationships for Text-Video Retrieval (AAAI 2025 · similarity 0.84)
- INQUIRE: A Natural World Text-to-Image Retrieval Benchmark (NeurIPS 2024 · similarity 0.84)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 2025 · similarity 0.83)
- Bridging Information Asymmetry in Text-video Retrieval: A Data-centric Approach (ICLR 2025 · similarity 0.83)
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.
