Research map / Video / Video-language understanding
Revisiting Temporal Modeling for CLIP-Based Image-to-Video Knowledge Transferring
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 STAN with branch-based multi-level temporal modeling instead of the CLIP baseline | CLIP · MSR-VTT | R@1: 43.1 -> 46.9 | Helped |
| Use STAN with branch-based multi-level temporal modeling instead of the CLIP baseline | CLIP · DiDeMo | R@1: 43.4 -> 46.2 | Helped |
| Use STAN with branch-based multi-level temporal modeling instead of the CLIP baseline | CLIP · Kinetics-400 | Top-1 accuracy: 79.9 -> 82.6 | Helped |
| Use STAN with branch-based multi-level temporal modeling instead of the CLIP baseline | CLIP · Something-Something-v2 | Top-1 accuracy: 44 -> 65.9 | Helped |
| Add the cross-frame temporal module (ablation) | STAN with intra-frame module and branch structure · Something-Something-v2 | Top-1 accuracy: 58.6 -> 65.9 | Helped |
| Add the intra-frame module (ablation) | STAN with cross-frame module and branch structure · Something-Something-v2 | Top-1 accuracy: 55.7 -> 65.9 | Helped |
| Use the branch structure and multi-level inputs rather than a posterior structure without multi-level inputs (ablation) | STAN · MSR-VTT | R@1: 44.2 -> 46.9 | Helped |
| Use the branch structure and multi-level inputs rather than a posterior structure without multi-level inputs (ablation) | STAN · Something-Something-v2 | Top-1 accuracy: 58.6 -> 65.9 | Helped |
| Select CLIP layer inputs at interval 1 instead of interval 2 (ablation) | STAN · MSR-VTT | R@1: 44.2 -> 46.9 | Helped |
| Select CLIP layer inputs at interval 1 instead of interval 2 (ablation) | STAN · Something-Something-v2 | Top-1 accuracy: 61.1 -> 65.2 | Helped |
| Use CLIP layers 9-12 rather than layers 1-4 as STAN inputs (ablation) | STAN · MSR-VTT | R@1: 43.4 -> 46.9 | Helped |
| Use CLIP layers 9-12 rather than layers 1-4 as STAN inputs (ablation) | STAN · Something-Something-v2 | Top-1 accuracy: 62.2 -> 65.2 | Helped |
Closest papers by meaning
- Temporal Reasoning Transfer from Text to Video (ICLR 2025 · similarity 0.88)
- VideoRefer Suite: Advancing Spatial-Temporal Object Understanding with Video LLM (CVPR 2025 · similarity 0.88)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.88)
- Seq2Time: Sequential Knowledge Transfer for Video LLM Temporal Grounding (CVPR 2025 · similarity 0.88)
- Enhanced Motion-Text Alignment for Image-to-Video Transfer Learning (CVPR 2024 · similarity 0.87)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · 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.