What it does

This is a curated awesome-list that catalogs popular deep learning models for classification, segmentation, and detection. Each entry includes parameter counts, FLOPS, and top-1/top-5 error rates pulled from the original papers. Think of it as a cheat sheet for the last decade of computer vision architecture trends.

The interesting bit

The value isn’t the curation itself—it’s the side-by-side comparison. You can watch ResNet-50 (25.5M params, 22.28% top-1 error) give way to EfficientNet variants, or see how SqueezeNet tried to murder AlexNet with 50× fewer parameters. The README is essentially a CSV that learned Markdown.

Key highlights

• Coverage spans 2012 (AlexNet) through recent efficient architectures (MobileNet, ShuffleNet, EfficientNet variants)
• Segmentation and detection sections include Mask R-CNN, YOLO variants, SSD, and RetinaNet with mAP scores
• Direct arXiv links for every model—no hunting for “which paper was that?”
• Includes niche entries like DiracNet (skip-connection-free training) and PolyNet (structural diversity obsession)
• DarkNet variants listed, because someone still needs to compare YOLO backbones

Caveats

• Several FLOPS cells show “?"—the README doesn’t explain whether these were unavailable or omitted
• No code implementations; this is purely a reference table, not a model zoo
• Last significant update timing is unclear from the README

Verdict

Grab this if you’re writing a paper related work section or picking a baseline and need to justify “we chose X because Y has fewer FLOPS.” Skip it if you want runnable code or training scripts—this is strictly a lookup table with links.