Finetuning a pretrained face model

vision
1

Hi,
I am trying to finetune a pretrained face model. I have used VGG face model in the below link.
http://www.robots.ox.ac.uk/~albanie/pytorch-models.html

But when i load the pretrained weights, it throws the below mentioned error.

RuntimeError: Error(s) in loading state_dict for Vgg_m_face_bn_dag:
size mismatch for fc7.weight: copying a param of torch.Size([4096, 4096, 1, 1]) from checkpoint, where the shape is torch.Size([4096, 4096]) in current model.

Can you please help me to fix the error.

Any suggestion to finetune a pretrained face model will be greatly appreciated

2

same problem! hope to get an answer.

3

unnamed

Try to replace fc7 and fc8 with linear operations. Maybe OK for extracting features without using fc layers. But not sure whether will work for finetuning.

Some says that this may due to the the pytorch version problem. What version do you use? Will old pytorch version (0.2.x) work?

Another problem: why the forward() return data in http://www.robots.ox.ac.uk/~albanie/models/pytorch-mcn/vgg_face_dag.py?