Research map / Video / Video segmentation and tracking
Color When It Counts: Grayscale-Guided Online Triggering for Always-On Streaming Video Sensing
Weitong Cai, Hang Zhang, Yukai Huang, Shitong Sun, Jiankang Deng, Songcen Xu, Jifei Song, Zhensong Zhang
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 |
|---|---|---|---|
| keep continuous grayscale stream alongside triggered RGB (vs Only TrigRGB, 8.1% RGB) (ablation) | ColorTrigger · StreamingBench (real-time visual understanding) | overall accuracy: 68.76 -> 70.72 | Improved |
| use triggered instead of uniform RGB frames (8.1% RGB) (ablation) | ColorTrigger + grayscale · StreamingBench (real-time visual understanding) | overall accuracy: 69.00 -> 70.72 | Improved |
| keep grayscale stream alongside triggered RGB (34.3% RGB) (ablation) | ColorTrigger · StreamingBench (real-time visual understanding) | overall accuracy: 73.48 -> 75.24 | Improved |
| use triggered instead of uniform RGB frames (34.3% RGB) (ablation) | ColorTrigger + grayscale · StreamingBench (real-time visual understanding) | overall accuracy: 72.64 -> 75.24 | Improved |
| add grayscale stream to uniform RGB sampling (8.1% RGB) (ablation) | Uniform RGB · StreamingBench (real-time visual understanding) | overall accuracy: 65.52 -> 69.00 | Improved |
| add dynamic token router (64 gray tokens) at 8.1% RGB (ablation) | Gray + TrigRGB · StreamingBench (real-time visual understanding) | overall accuracy: 67.08 -> 70.72 | Improved |
| add dynamic token router (64 gray tokens) at 34.3% RGB (ablation) | Gray + TrigRGB · StreamingBench (real-time visual understanding) | overall accuracy: 68.96 -> 75.24 | Improved |
| use 256 gray tokens instead of dynamic router at 34.3% RGB (ablation) | Gray + TrigRGB · StreamingBench (real-time visual understanding) | overall accuracy: 75.24 -> 76.32 | Mixed |
| insert 8.1% triggered RGB frames into grayscale video (ablation) | InternVL-3.5-8B grayscale · StreamingBench (real-time visual understanding) | overall accuracy: 62.08 -> 70.72 | Improved |
| insert 33.1% triggered RGB frames into grayscale video | InternVL-3.5-8B grayscale · OVO-Bench | overall average: 47.9 -> 52.5 | Improved |
| use 37.6% triggered RGB instead of full RGB | InternVL-3.5-8B · Video-MME | overall accuracy: 65.6 -> 66.1 | Improved |
| replace full RGB by 34.3% triggered RGB | InternVL-3.5-8B · StreamingBench (real-time visual understanding) | overall accuracy: 77.20 -> 75.24 | Hurt |
Closest papers by meaning
- E-RGB-D: Real-Time Event-Based Perception with Structured Light (IJCV 2026 · similarity 0.84)
- Tracking Emerges by Colorizing Videos (ECCV 2018 · similarity 0.83)
- RGB-Event ISP: The Dataset and Benchmark (ICLR 2025 · similarity 0.83)
- Adaptive Depth Lightweight RGB-T Tracking with Holistic Token Routing (CVPR 2026 · similarity 0.83)
- ColorFM: An Optimization-to-Learning Framework for Color Transfer via Flow Matching (ECCV 2026 · similarity 0.83)
- ISPDiffuser: Learning RAW-to-sRGB Mappings with Texture-Aware Diffusion Models and Histogram-Guided Color Consistency (AAAI 2025 · similarity 0.82)
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.
