joonson / yousaidthat Goto Github PK

Hey I am trying to run this repo on Windows with CUDA8, but when I run "vl_compilenn('enableGpu', true, ...
'cudaRoot', 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0', ...
'cudaMethod', 'nvcc') ;"
I get the following error

@joonson @amirjamaludin
Hi, Thank you for your contribution on this work.
I'm interested in this work, and I'd like to reproduce it in Tensorflow.
In section 2 of your paper, you said that 37.7 hours, 678,389 samples from both VoxCeleb and LRW datasets are used for training.
When preparing data, I found videos far more than 37.7 hours.
So, how did you filter the dataset to 37.7 hours?
Besides, how many hours did you use for training?
Thank you!
I'm looking forward to your reply!

First thank you for your nice idea for generating the talking face form a unknow audio.When I try to run your demo, I got wrong answer and i have no idea what's the problem.Here is the problem.

    mfcc = gpuArray(repmat(single(C (2:13,j:j+34)),1,1,1,numface));
    net.eval({'input_audio',mfcc,'input_face',faceY});
    im = gather(net.getVar('prediction').value);
    netR.eval({'input_lip',gpuArray(im)});
    imb         = gather(netR.vars(end).value);
    imR3        = imtile(imresize((im+imb),[109 109]),numcol,numrow);
    imO3        = imtile(imresize(im,[109 109]),numcol,numrow);

No matter what the mfcc and faceY is , I got the im all equal to 0.5, and got the imb all equal to 0.0, so in the end i got the imR3 all equal to 0.5. At last I got a image with full 0.5 value pixel which result in a gray pic.Here is the result

Could you please give me some advice? I havent change any codes in the demo . Thank you very much.

What does these fields mean in Voxceleb2 header?

	Offset    : 	-2
	FV Conf   : 	16.303	(1)
	ASD Conf  : 	6.201