biubug6 / Pytorch_Retinaface

Retinaface get 80.99% in widerface hard val using mobilenet0.25.
MIT License
2.63k stars 774 forks source link

Model does not converge when i change the input size from 640*640 to 360*640. #90

Open ThomsonW opened 4 years ago

ThomsonW commented 4 years ago

I change the input size of images to 360*640, and then re-train the model. But the model dose not converge. I have checked the part of data augmentation and there is no problem. Also, I check multibox_loss.py and don't find any setting about input size. Intuitively, such change may not lead the problem of convergence. So is there something i missing or else? thanks very much.