Research map / Video / Video segmentation and tracking
Video-Text Representation Learning via Differentiable Weak Temporal Alignment
Dohwan Ko, Joonmyung Choi, Juyeon Ko, Shinyeong Noh, Kyoung-Woon On, Eun-Sol Kim, Hyunwoo J. Kim
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 |
|---|---|---|---|
| Use VT-TWINS instead of MIL-NCE with S3D | video representations · HMDB | accuracy: 53.1 -> 57.9 | Improved |
| Use VT-TWINS instead of MIL-NCE with S3D | video representations · UCF | accuracy: 82.7 -> 85 | Improved |
| Use VT-TWINS instead of the strongest listed baseline | video-text representations · YouCook2 text-to-video retrieval | R@1: 9.6 -> 9.7 | Improved |
| Use VT-TWINS instead of MIL-NCE | video-text representations · MSR-VTT text-to-video retrieval | R@1: 8.2 -> 9.4 | Improved |
| Use VT-TWINS instead of reproduced MIL-NCE | video-text representations · YouCook2 video-to-text retrieval | R@10: 37.36 -> 40.3 | Improved |
| Use VT-TWINS instead of reproduced MIL-NCE | video-text representations · MSR-VTT video-to-text retrieval | R@10: 27.2 -> 29.1 | Improved |
| Use VT-TWINS instead of ActBERT | action-step localization representations · CrossTask | CTR: 37.1 -> 40.7 | Improved |
| Add temporal augmentation that suppresses semantically strong permutations (ablation) | standard-DTW contrastive learning · YouCook2 | R@10: 8.7 -> 9.6 | Improved |
| Use random temporal permutations instead of no temporal augmentation (ablation) | standard-DTW contrastive learning · YouCook2 | R@10: 8.7 -> 5 | Hurt |
| Encourage semantically strong temporal permutations instead of no temporal augmentation (ablation) | standard-DTW contrastive learning · YouCook2 | R@10: 8.7 -> 4.9 | Hurt |
| Add weak alignment to temporal augmentation A (ablation) | standard-DTW contrastive learning · YouCook2 | R@10: 9.6 -> 10.6 | Improved |
| Add local neighborhood smoothing to augmentation A and weak alignment (ablation) | standard-DTW contrastive learning · YouCook2 | R@10: 10.6 -> 12.5 | 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.
- ActBERT — ActBERT: Learning Global-Local Video-Text Representations (CVPR 2020) · better on 3 of 3 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
- Learning by Aligning Videos in Time (CVPR 2021 · similarity 0.89)
- Dynamic Reflections: Probing Video Representations with Text Alignment (ICLR 2026 · similarity 0.88)
- Weakly Supervised Video Representation Learning With Unaligned Text for Sequential Videos (CVPR 2023 · similarity 0.88)
- TempCLR: Temporal Alignment Representation with Contrastive Learning (ICLR 2023 · similarity 0.88)
- ITO: Multi-View Alignment and Training-Time Fusion for Image-Text Pretraining (NeurIPS 2026 · similarity 0.87)
- VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models (CVPR 2025 · similarity 0.86)
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.
