A field guide to deep-learning image fusion

What it does This repository functions as a curated field map for deep-learning image fusion research. It sorts hundreds of papers into taxonomies—multi-modal (infrared/visible, medical), digital photography (multi-exposure, multi-focus), remote sensing (pansharpening), and general frameworks—tagging each by base architecture (CNN, Transformer, Diffusion, autoencoder) and supervision style. It also serves as the companion hub for the authors’ own survey paper in the Journal of Image and Graphics.

The interesting bit The taxonomy captures the field’s splintering goals: some methods chase low-level visual quality, others inject semantic priors from detection or segmentation tasks, and a newer branch tackles joint registration and fusion for misaligned inputs. The README doubles as a research radar, tracking the authors’ prolific output across CVPR, NeurIPS, TPAMI, and IJCV.

Key highlights

• Annotated tables linking to papers and code for hundreds of methods
• Classification by framework (CNN, GAN, Transformer, Diffusion) and supervision paradigm
• Coverage of niche branches like degradation-robust and language-guided fusion
• Pointers to evaluation metrics and datasets
• Companion to a survey paper that has won multiple journal excellence awards

Caveats

• The repository is primarily a curated index and survey companion; most runnable implementations live in separate repositories linked from the tables.
• The README is predominantly Chinese with English paper titles and section headers.

Verdict Researchers entering the image-fusion literature should bookmark this as a navigational anchor. Practitioners looking for a single, drop-in fusion library will need to follow the outbound links to specific method repositories.