AI Paper Atlas

Research map / Image generation / Image restoration

HiDRA: Hierarchical Degradation Representation and Adaptation with Generative Priors for Enhancing Infrared Vision

Zihang Chen, Zhu Liu, Changbo Yan, Jinyuan Liu, Risheng Liu

CVPR 2026 · Image generation › Image restoration · approach: Efficient restoration architectures

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
use the bi-level first-order solution instead of alternating optimization (ablation)HiDRA · HM-TIR blind super-resolutionMANIQA: 0.530 -> 0.556Improved
use the bi-level first-order solution instead of alternating optimization (ablation)HiDRA · HM-TIR FPN correctionMANIQA: 0.556 -> 0.572Improved
use upper-level bi-level training instead of joint training of DEM and base model (ablation)HiDRA · HM-TIR FPN correctionMANIQA: 0.569 -> 0.572Improved
use upper-level bi-level training instead of joint training of DEM and base model (ablation)HiDRA · HM-TIR blind super-resolutionMANIQA: 0.556 -> 0.556No clear effect
bi-level learning makes DEM embeddings separable by degradation level (ablation)DEM · t-SNE of degradation vectorscluster separabilityImproved
increase number of degradation pipelines M from 2 to 3 (ablation)HiDRA · HM-TIR blind SR, 256x256 patchesMANIQA: 0.500 -> 0.548Improved
increase number of degradation pipelines M from 3 to 4 (ablation)HiDRA · HM-TIR blind SR, 256x256 patchesMANIQA: 0.548 -> 0.543Hurt
add DEM with MIA module to LoRA fine-tuning (ablation)SD Turbo + LoRA · HM-TIR (Fig. 6a)image quality metricsImproved
replace deterministic composite-degradation framework (PPFN) with HiDRA at extreme degradationHiDRA · Composite degradation (Extreme)LPIPS: 0.550 -> 0.230Improved
use HiDRA enhancement before downstream detectorYOLOv8 on M3FD · M3FD object detectionmAP: 0.595 -> 0.619Improved
use HiDRA super-resolution before segmentationSegFormer on FMB · FMB semantic segmentationmIoU: 48.2 -> 49.4Improved
HiDRA versus PPFN on real-world no-reference fusion metricsHiDRA · TNO / RoadScene real-worldQAB/F: 0.3916 -> 0.3912 (TNO)Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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