The Receipt for Computer Vision's Transformer Obsession

What it does

This is a curated bibliography—an “awesome list” in the classic sense—collecting recent research papers that apply Transformer architectures to computer vision. The repository organizes papers into roughly 80 task-specific categories, from backbone classification and object detection to niche domains like hyperspectral imaging, gait recognition, and neural rendering. Each section links to the paper and, where available, its accompanying code.

The interesting bit

The taxonomic ambition is the real story. The list reads like a longitudinal study of a field-wide paradigm shift, documenting that Transformers are no longer just an alternative to convolutions for image recognition but the default tool for dehazing, matting, SLAM, and even model pruning. It is maintained like a community wiki, accepting pull requests for overlooked papers.

Key highlights

• Covers roughly 80 distinct CV tasks and subdomains, including medical imaging, remote sensing, point clouds, and diffusion models
• Links to code implementations when authors have made them available
• Maintains archived snapshots (e.g., list_old_20240323.md) to preserve historical states of the bibliography
• Accepts community contributions via issues and pull requests for missing papers

Caveats

• Strictly a reading list with no executable code or frameworks
• Paper entries reside in separate markdown files under main/, requiring click-through browsing
• Inclusion aims for comprehensiveness, not curation; no quality ranking is provided

Verdict

Researchers and review-paper writers drowning in Transformer CV literature should bookmark this. If you need training scripts or pretrained weights, look elsewhere.