A detailed and optimized implementation with full featured Yolo-V3 advantages.
Yolo-V3 has been updated head network structure (DarkNet 19 -> DarkNet 53). This improvement was inspired by ResNet and FPN. Yolo-V3 has very good improvements then previous version. As well as accelerate the speed of detection, roughly 3-4 times then SSD or Retina-Net.
To train our own datasets, we should prepared some raw labels, I will public some articles about this. Currently we are structing the training pipleline codes.
Todo.
This work was done by Jin Fagang, All Rights Reserved.