What it does

This is a curated learning path through deep learning for image processing, built by a graduate student documenting their own research journey. Each topic gets a theory video, a PyTorch implementation walkthrough, and often a TensorFlow/Keras equivalent. Course PPTs live in course_ppt; code presumably lives elsewhere in the repo.

The interesting bit

The dual-framework mandate is the quietly useful part. Most tutorials pick PyTorch or TensorFlow and call it a day; this one forces both, which is annoying for the author and genuinely educational for anyone who has to read codebases in either. The coverage is also remarkably current—Swin Transformer, ConvNeXt, MobileViT, YOLOX all made the cut.

Key highlights

• Covers classification (LeNet through MobileViT), detection (Faster R-CNN through YOLOX), segmentation (FCN, DeepLab, U-Net, U2Net), instance segmentation (Mask R-CNN), and keypoint detection (DeepPose, HRNet)
• Every major architecture gets: network architecture explanation → PyTorch build → TensorFlow2 build
• All videos hosted on Bilibili; PPTs downloadable from repo
• Maintained as an ongoing project with new architectures added post-graduation
• 26K+ stars suggests it’s filling a real gap in Chinese-language CV education

Caveats

• README is essentially a video index; code visibility and quality are unclear without digging past the landing page
• TensorFlow coverage appears slightly spotty (some entries missing the TF2 link, or using inconsistent naming like “Tensorflow” vs “Tensorflow2”)
• No English translation apparent; accessibility limited to Chinese speakers

Verdict

Worth bookmarking if you’re learning CV in Chinese or need to cross-train between PyTorch and TensorFlow. Skip if you want production-ready libraries or English-language documentation.