December 2020
tl;dr: Pytorch-native improved version of Yolov4.
The author of the repo has not released a paper yet, but the repo is highly useful. Many Kaggler commented that its performance is better than yolov4. The training speed of Yolov5 is also much faster than Yolov4.
- Two great reviews on Zhihu
- Focus layer
- This is a spatial to channel layer that warps H x W x 3 into H/2 x W/2 x12.
- See issue in yolov5 github
- Papers such as TResNet WACV 2021 and Isometric Neural Networks ICCV 2019 workshop
- Adaptive anchor learning with genetic algorithm
- Mosaic data aug was first invented in ultralytics's yolov3 and borrowed into Yolov4.
- Questions and notes on how to improve/revise the current work