Research map / Video / Video-language understanding
Prototypes are Balanced Units for Efficient and Effective Partially Relevant Video Retrieval
WonJun Moon, Cheol-Ho Cho, Woojin Jun, Taeoh Kim, Inwoong Lee, Dongyoon Wee, Minho Shim, Jae-Pil Heo
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 cross-modal mask reconstruction (ablation) | prototype generator · TVR (CLIP L/14) | SumR: 253.0 -> 254.2 | Improved |
| add uni-modal reconstruction (ablation) | prototype model with cross-modal reconstruction · TVR (CLIP L/14) | SumR: 254.2 -> 255.1 | Improved |
| add weak attention guidance via video mixing (ablation) | prototype model with dual reconstruction · TVR (CLIP L/14) | SumR: 255.1 -> 258.2 | Improved |
| add orthogonal prototype diversification loss (ablation) | prototype model with guidance · TVR (CLIP L/14) | SumR: 258.2 -> 259.2 | Improved |
| use 10 prototypes instead of 30 (ablation) | Ours · TVR (CLIP L/14) | SumR: 259.2 -> 252.9 | Hurt |
| use 20 prototypes instead of 30 (ablation) | Ours · TVR (CLIP L/14) | SumR: 259.2 -> 257.8 | Hurt |
| use 60 prototypes instead of 30 (ablation) | Ours · TVR (CLIP L/14) | SumR: 259.2 -> 260.2 | Improved |
| replace cross-attention-only generator with Perceiver (ablation) | Ours · TVR (CLIP L/14) | SumR: 259.2 -> 249.5 | Hurt |
| replace cross-attention-only generator with Slot-attention (ablation) | Ours · TVR (CLIP L/14) | SumR: 259.2 -> 253.8 | Hurt |
| use prototypes instead of GMMFormer local-clip modeling (ablation) | GMMFormer · TVR (CLIP L/14) | SumR: 243.2 -> 259.2 | Improved |
| use prototypes versus storing all clips (MS-SL) (ablation) | MS-SL · TVR (CLIP L/14) | Memory (MB): 1.01 -> 0.09 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GMMFormer — GMMFormer: Gaussian-Mixture-Model Based Transformer for Efficient Partially Relevant Video Retrieval (AAAI 2024) · better on 8 of 10 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Bridging the Semantic Granularity Gap Between Text and Frame Representations for Partially Relevant Video Retrieval (AAAI 2025 · similarity 0.91)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 2025 · similarity 0.91)
- Ambiguity-Restrained Text-Video Representation Learning for Partially Relevant Video Retrieval (AAAI 2025 · similarity 0.88)
- Video-ColBERT: Contextualized Late Interaction for Text-to-Video Retrieval (CVPR 2025 · similarity 0.88)
- Generative Video Diffusion for Unseen Novel Semantic Video Moment Retrieval (AAAI 2025 · similarity 0.87)
- Composed Video Retrieval via Enriched Context and Discriminative Embeddings (CVPR 2024 · 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.
