Sample Dataset for usage #2

[shikhar-scs]

Hi @Rudrabha ,
Amazing work.
I was working on generating video from image+audio and it would be very helpful if you could post a sample image and audio file. I've been getting different errors every time I'm using a random image.

Thanks!

[prajwalkr]

Thank you very much for your interest.

Could you please some of the errors? If it is a mistake in the code, then we can fix it for everyone.

[shikhar-scs]

Hey, I did a setup of the same on local(mac) and it worked fine.

On remote (gpus) there was the following tensorflow error

[prajwalkr]

The input size must be 96x96. Please ensure this.

If it doesn't get solved with this, please report the tensorflow and keras version you have.

[prajwalkr]

The input size must be 96x96. Please ensure this.

Have updated the repo code as well to not receive img_size as a variable input parameter.

[shikhar-scs]

Oh okay, thanks for the help !

[shikhar-scs]

hey @prajwalkr , another thing.

For training the model could you please specify the dataset structure, would be helpful for a lot of people, the way its mentioned here https://github.com/Hangz-nju-cuhk/Talking-Face-Generation-DAVS#preparing-training-data

[prajwalkr]

Done.

[shikhar-scs]

Hi @Rudrabha finally completed the end to end setup and more importantly everything is working now.
Just a few nit bugs in preprocess.py. Pointing out so that future users don't have to spend extra time.

The split in line 116 & 118 should probably be args.split

LipGAN/preprocess.py

Lines 116 to 118 in 16ca935

	for line in open(path.join(args.filelists, '{}.txt'.format(split))).readlines()]
	jobs = [(vfile, args, split) for vfile in filelist]

Also, sr here, the sampling rate, is undefined again. Is there any specific value to be used ? (as of now I've set it to None, to preserve the native sampling rate of the file).

LipGAN/preprocess.py

Line 87 in 16ca935

wav = audio.load_wav(wavpath, sr)

Will let you know if I come across any others. Thanks again for the work!

[prajwalkr]

Please use sr=16000, I will correct these two in preprocess.py right away. Thank you very much.