search

1adrianb / face-alignment-training

Training code for the networks described in "How far are we from solving the 2D & 3D Face Alignment problem? (and a dataset of 230,000 3D facial landmarks)" paper.
https://www.adrianbulat.com/face-alignment
Other
144 stars 48 forks source link

I have a problem for caffe implementation #2

Closed anshan-XR-ROB closed 6 years ago

anshan-XR-ROB commented 6 years ago

Hi, I have a problem for caffe implementation. table.insert(out, tmpOut) I want to know the meaning of this line. I only use tmpOut4 for output and the result is poor. Shoud l concat all tmpOut to output?

1adrianb commented 6 years ago

Hi @AnshanTJU,

That particular line puts each intermediate output in a table which corresponds with the number of stacked modules (in this case 4) . When training you need to computer the loss against each of this intermediate outputs (so the loss is a linear combination of all them). When evaluating the last output should give you the best results. Addionally it is possible that some implementations of the optimisation methods (ex: rmsprop) or intialisation may behave different between torch and caffe so you may need to adjust this to match the results during training.