AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

PhysDNet: Physics-Driven Gradient Amplification for Real-Time Image Dehazing

Rayen Dhahri, Lei Sun, Victor J Morinigo, Alice Pagnoux, Luca Benini, Danda Pani Paudel, Luc V Gool

NeurIPS 2026 · Architectures and efficiency › RNNs and normalization · approach: Dataset condensation and flow estimation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Introduces PhysDNet, a lightweight multimodal dehazing network that fuses NIR imagery with sparse LiDAR depth for real-time restoration on edge devices.

Problem

Modern dehazing networks are computationally expensive and overfit to synthetic haze, limiting real-time deployment on resource-constrained edge devices.

Approach

A lightweight architecture that fuses NIR imagery with sparse LiDAR depth, with physics-driven gradient amplification; the abstract is truncated so further training details are not available.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add physics head (Koschmieder inversion) as auxiliary training head (ablation)PhysDNet-M · STF valPSNR: 30.02 -> 35.55Improved
add physics head (ablation)PhysDNet-M · STF YOLOv8 detection (fog 0.06)mAP: 10.70 -> 11.70Improved
add physics head (ablation)PhysDNet-M · O-HAZE zero-shotPSNR-Y: 18.92 -> 20.45Improved
add physics head (ablation)PhysDNet-M · I-HAZE zero-shotPSNR-Y: 16.33 -> 16.93Improved
add physics head when no transmission anchor (RGB-only, no depth) (ablation)PhysDNet-M · SOTS-mix (RESIDE-6K)PSNR: 28.6 -> 28.4No clear effect
use PhysDNet-L over FFA-Net+LiDAR 512dehazing network · STF valPSNR: 34.79 -> 36.46Improved
add LiDAR input to DehazeFormer-T / AOD-Net at 512baselines · STF valPSNR: 33.38 -> 29.91 (DehazeFormer-T 512)Hurt
add physics head to NH-HAZE non-homogeneous haze (ablation)PhysDNet-M · NH-HAZEPSNR-Y: -0.54 dB without physics headImproved

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