Open xuezu29 opened 2 years ago
Yes, we have developed some anchor free methods. It is useful for tiny models and small input size, but for large input size, anchor based methods are still easier converged.
@WongKinYiu That would explain some of the problems with YOLOX's training, when I increase the size of the input image, The results may not be better, thanks a lot!
First, Thank you for your team's research and contribution! Have you tested the results of the free anchor like YOLOX, thx!