AI Paper Atlas

Research map / Image generation / Image restoration

InstructIR: High-Quality Image Restoration Following Human Instructions

Marcos V. Conde, Gregor Geigle, Radu Timofte

ECCV 2024 · Image generation › Image restoration · approach: Efficient restoration architectures · 145 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
add human-written text instructions to NAFNet restoration (ablation)InstructIR · 5-task averagePSNR: 27.84 -> 29.55Improved
add text instructions on GoPro deblurring (ablation)InstructIR · GoProPSNR: 26.65 -> 29.40Improved
add text instructions in the 3-task setting (ablation)InstructIR · 3D averagePSNR: 30.65 -> 32.43Improved
use ambiguous instead of precise instructions (ablation)InstructIR-5D · LOLPSNR: 23.00 -> 21.85Hurt
use ambiguous instead of precise instructions (ablation)InstructIR-5D · Rain100LPSNR: 36.84 -> 36.24Hurt
swap text encoder BGE-micro for all-MiniLM or CLIP (ablation)InstructIR-5D · LOLPSNR: 23.00 -> 22.98 (MiniLM), 22.95 (CLIP)No clear effect
add super-resolution as sixth training task (ablation)InstructIR-5D · GoProPSNR: 29.40 -> 29.73Improved
add MIT5K enhancement as seventh task (ablation)InstructIR-6D · Rain100LPSNR: 36.80 -> 36.75No clear effect
train on 5 tasks instead of 3 tasks (ablation)InstructIR · Rain100LPSNR: 37.98 -> 36.84Hurt
use InstructIR with 16M parameters instead of PromptIRall-in-one restoration · Inference costMACs (G): 160 -> 100Improved

Later papers that compared with this method

8 later accepted papers report InstructIR as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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