Research map / Vision-language and multimodal / CLIP and zero-shot
TransVLM: A Vision-Language Framework and Benchmark for Detecting Any Shot Transitions
Ce Chen, Yi Ren, Yuanming Li, Viktor Goriachko, Zhenhui Ye, Zujin Guo, Zhibin Hong, Mingming Gong
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 optical flow input fused with color frames (ablation) | TransVLM · STD public data | segment F1: 0.694 -> 0.783 | Improved |
| add optical flow input (ablation) | TransVLM · STD synthetic data | frame F1: 0.892 -> 0.938 | Improved |
| fuse flow at patch embedding instead of separate inputs (ablation) | TransVLM · STD benchmark | RTF: 1.29 -> 0.50 | Improved |
| fuse flow at patch embedding instead of separate inputs (ablation) | TransVLM · STD public data | segment F1: 0.739 -> 0.783 | Improved |
| zero-padding init instead of duplicating color weights (ablation) | TransVLM · STD public data | segment F1: 0.655 -> 0.783 | Improved |
| add fine-tuning on STD training data (ablation) | Qwen3-VL-4B · STD public data | segment F1: 0.487 -> 0.783 | Improved |
| add public data to synthetic-only training (ablation) | TransVLM · STD public data | segment F1: 0.416 -> 0.783 | Improved |
| add synthetic data to public-only training (ablation) | TransVLM · STD synthetic data | segment F1: 0.720 -> 0.895 | Improved |
| add synthetic data to public-only training (ablation) | TransVLM · STD public data | segment F1: 0.730 -> 0.783 | Improved |
| add public data to synthetic-only training (ablation) | TransVLM · STD synthetic data | frame F1: 0.941 -> 0.938 | No clear effect |
| use TransVLM instead of Qwen3-VL-4B Instruct zero-shot | Qwen3-VL-4B · STD public data | segment F1: 0.124 -> 0.783 | Improved |
| TransVLM vs AutoShot on boundary error and speed | TransVLM · STD public data | ABE (s): 1.58 vs 1.78 ABE; RTF 0.50 vs 0.03 | Hurt |
Closest papers by meaning
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.87)
- Towards Understanding Camera Motions in Any Video (NeurIPS 2025 · similarity 0.87)
- ShotBench: Expert-Level Cinematic Understanding in Vision-Language Models (NeurIPS 2025 · similarity 0.86)
- RE-VLM: Event-Augmented Vision-Language Model for Scene Understanding (CVPR 2026 · similarity 0.86)
- Dense Video Object Captioning from Disjoint Supervision (ICLR 2025 · similarity 0.86)
- Modality-Aware Shot Relating and Comparing for Video Scene Detection (AAAI 2025 · similarity 0.85)
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.
