AI Paper Atlas

Research map / Detection and segmentation / CNN features

Beyond RGB: Adaptive Parallel Processing for RAW Object Detection

Shani Gamrian, Hila Barel, Feiran Li, Masakazu Yoshimura, Daisuke Iso

ICCV 2025 · Detection and segmentation › CNN features · approach: Context for hard detection conditions · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
apply ISP functions in parallel with shared encoder instead of sequentially (ablation)RAM · LOD-DarkmAP: 29.9 -> 35.2Improved
apply ISP functions in parallel instead of sequentially (ablation)RAM · LOD-NormalmAP: 35.1 -> 40.1Improved
replace sRGB ISP with RAM on RAW inputFaster R-CNN ResNet18 · ROD-NightmAP: 38.9 -> 44.5Improved
scale backbone from ResNet18 to ResNet50 on RAW (ablation)Faster R-CNN on RAW · ROD-NightmAP: 30.4 -> 34.7Improved
use tiny RAM-T with fewer than half the parameters (ablation)RAM · ROD-NightmAP: 44.5 -> 44.2No clear effect
apply RAM to sRGB images instead of RAW (ablation)RAM · LOD-DarkmAP: 34.9 -> 32.9Hurt
use RAM on noisy RAW instead of sRGBDINO ResNet50 · ROD-Night strong noisemAP: 38.9 -> 42.4Improved
use RAM under snow weather instead of sRGBDINO ResNet50 · ROD-Night snowmAP: 37.4 -> 53.2Improved
train RAM only with frozen COCO-pretrained YOLOX-X detectorRAM · NOD-SonymAP: 24.4 -> 29.6Improved
train one model on all four RAW datasets instead of each separately (ablation)RAM · NOD-SonymAP: 32.8 -> 35.8Improved
add brightness adjustment function (ablation)RAM · LOD-NormalShapley valueNo clear effect
use reverse-hourglass Feature Fusion instead of single-layer convolution (ablation)RAM · multiple RAW datasetsShapley valueImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy