Hi @1adrianb , thank you for your great work, it is really cool.
However, I met some problems when I try to reproduce 2d-fan result by pytorch. I wrote my training process inspired by 2d-fan code in lua version . https://github.com/1adrianb/2D-and-3D-face-alignment
And my training loss is MSE. My loss is decreased to around 400 with a bad performance. However i found the model you provided have a very good performance with MSE loss is around 1000.
It is strange, right? Maybe you can give me some advice. Is MSE suitable in this problem?
Hi @1adrianb , thank you for your great work, it is really cool. However, I met some problems when I try to reproduce 2d-fan result by pytorch. I wrote my training process inspired by 2d-fan code in lua version . https://github.com/1adrianb/2D-and-3D-face-alignment And my training loss is MSE. My loss is decreased to around 400 with a bad performance. However i found the model you provided have a very good performance with MSE loss is around 1000. It is strange, right? Maybe you can give me some advice. Is MSE suitable in this problem?