Visual TL;DR
VLMs use RGBDriver
From the articleVision-language models (VLMs) typically operate on post-image signal processing (ISP) RGB images.
RGB loses dataDriver
preprocessing discards crucial sensor evidence through clipping, suppression, or quantization
From the articleThis technique effectively transfers supervision signals from readily available RGB proxies to the more granular, raw measurement-domain observations, addressing a fundamental challenge in training on sensor data.
PRISM-VL approachCore
grounds vision-language models in raw camera measurements, not just RGB
From the article 2 mentionsA new approach, PRISM-VL, investigates whether grounding performance improves when the visual interface is moved closer to the original camera measurement.
RAW-derived Meas.-XYZContext
directly incorporates raw sensor data inputs for improved grounding
From the articleThis framework directly incorporates RAW-derived Meas.-XYZ inputs.
Camera-conditioned groundingCore
a key innovation for better understanding of sensor data
From the article 3 mentionsA key innovation is its camera-conditioned grounding mechanism and Exposure-Bracketed Supervision Aggregation.
Improved performanceOutcome
significantly improves performance on challenging visual tasks
From the articleA new approach, PRISM-VL, investigates whether grounding performance improves when the visual interface is moved closer to the original camera measurement.
Exposure-Bracketed SupervisionCore
transfers supervision from RGB proxies to raw measurement domain observations
From the article 2 mentionsA key innovation is its camera-conditioned grounding mechanism and Exposure-Bracketed Supervision Aggregation.
Quantifiable gainsOutcome
demonstrates measurable improvements in challenging scenarios
From the articleThis represents a substantial leap over the RGB-based Qwen3-VL-8B baseline, with gains of +0.1074 BLEU, +0.1071 ROUGE-L, and +4.46 percentage points in LLM-Judge accuracy.
