Research map / Video / Video-language understanding
Dynamic Reflections: Probing Video Representations with Text Alignment
Maks Ovsjanikov, Viorica Patraucean, Leonidas Guibas, Tyler Zhu, Tengda Han
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 |
|---|---|---|---|
| use multiple captions (1 to 10) at test time (ablation) | vision-text alignment · VaTeX | mutual kNN alignment: +60% from 1 to 10 captions | Improved |
| use more video frames at test time (ablation) | VideoMAEv2 · VaTeX | mutual kNN alignment | Improved |
| use more frames for image models (DINOv2) (ablation) | DINOv2 · PVD | mutual kNN alignment | No clear effect |
| average image features over 8 frames instead of single frame (ablation) | image models · VaTeX | mutual kNN alignment: 0.206 -> 0.223 (best) | Improved |
| use Gemma-2 text encoder instead of older text encoders (ablation) | text encoder · VaTeX | mutual kNN alignment: 0.18 -> 0.206 | Improved |
| add synthesized captions (Gemini 2.5 Pro) at test time (ablation) | vision models · PVD | mutual kNN alignment | Improved |
| increase synthesized captions beyond about 6 (ablation) | vision models · PVD | mutual kNN alignment | No clear effect |
| self-supervised video encoder (VideoMAEv2) instead of image encoder (DINOv2) | vision encoder · VaTeX | text alignment | Improved |
| higher text alignment vs downstream performance, SSL video models | SSL video models · SSv2, Kinetics, depth, pose, tracking | correlation with downstream | Mixed |
| use negative reordered captions instead of positive | text encoders · VideoComp temporal reorder | alignment | Hurt |
Closest papers by meaning
- VideoComp: Advancing Fine-Grained Compositional and Temporal Alignment in Video-Text Models (CVPR 2025 · similarity 0.91)
- Do Vision and Language Encoders Represent the World Similarly? (CVPR 2024 · similarity 0.90)
- Towards Balanced Alignment: Modal-Enhanced Semantic Modeling for Video Moment Retrieval (AAAI 2024 · similarity 0.90)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.90)
- VidLA: Video-Language Alignment at Scale (CVPR 2024 · similarity 0.89)
- Can Text-to-Video Generation help Video-Language Alignment? (CVPR 2025 · similarity 0.89)
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.
