Loss: nan with example WIDER FACES datatset

Author: TonyCongqianWangCreated Jun 27, 2024Updated Jun 27, 2024

I cloned your repository and used python 3.6 and pytorch 1.2. Still I get loss nan when I use the model to train on the widerfaces dataset you provide. I set learning rate to 1e-3 and even use the pre trained model as initialization

Source: Linzaer/Ultra-Light-Fast-Generic-Face-Detector-1MB