why pretrained 2D-FAN and 3D-FAN models are the same? about 2d-and-3d-face-alignment HOT 6 CLOSED

Hi,
What do you mean by the same? The files are different.
Please note however that the 3D model is predicting the 2D projections of the 3D points (without the depth).

If you are interested into predicting also the depth please have a look at: https://arxiv.org/pdf/1609.09545.pdf

If you have any other question or you experience issues with the code feel free to contact us.

from 2d-and-3d-face-alignment.

Hi 1adrianb:
Thank you for your quick reply. The same means all the weights value in the file are the same except the file name.

from 2d-and-3d-face-alignment.

@ForestWang
I double checked the models and they are different.
2D: https://www.adrianbulat.com/downloads/FaceAlignment/2D-FAN-300W.t7
3D: https://www.adrianbulat.com/downloads/FaceAlignment/3D-FAN.t7

Maybe you are somehow downloading the wrong ones?

Note that when you run main.lua you can specify the model that you want to load using -model [path_to_model]

Please let me know if everything is sorted out.

from 2d-and-3d-face-alignment.

Hi Adrian:
I downloaded the models again. The problem resolved. thank you very much.

from 2d-and-3d-face-alignment.

Hi Adrian:
I read your paper "two-stage", it's a wonderful work. in the paper, say that the source code is at http://www.cs.nott.ac.uk/~psxab5/

bu seems that it's not there already?
it's not open or at another link?

thanks very much.

from 2d-and-3d-face-alignment.

Hi, yes you are right, unfortunate at this moment there is no code publicly available. However all you have to do is for example to take the output of this network, concat it with the image and serve it as as input to a ResNet trained for depth prediction. You can train this altogether using the 3DFAW dataset.
This been said, I do plan to release them at later time during this summer.

from 2d-and-3d-face-alignment.