svishwa/crowdcount-mcnn

The model running result is different from README

Opened this issue · 1 comments

yin-yue-liang-21 commented

I am in the restoration experiment. At the beginning of model training, when epoch=0, MAE: 22.2 and MSE: 39.8. When epoch=2000, MAE: 11.7, MSE: 24.1. This is significantly different from the MAE=110 and MSE=169 in Shanghai Tech A mentioned in README. Have you also experienced this situation? Or did I ignore something?
1

2

yin-yue-liang-21 commented

oh no ，its training MAE, not predict MAE，i have try predict MAE ,its very same as README