ZiqiaoPeng / SyncTalk

[CVPR 2024] This is the official source for our paper "SyncTalk: The Devil is in the Synchronization for Talking Head Synthesis"
https://ziqiaopeng.github.io/synctalk/
Other
1.26k stars 149 forks source link

Trained on custom data still getting May Lady's face in my output. #166

Open HardikJain02 opened 4 months ago

HardikJain02 commented 4 months ago

I wonder why this lady's face is there?

Trained on this video: https://www.youtube.com/watch?v=tv-YUsK2Xsw [Removed yellow shirt guy from the video] Output video: https://github.com/ZiqiaoPeng/SyncTalk/assets/69311953/553de1f4-4931-4609-a60c-acf3485d7211

My model directory looks like this:

[Working on Goyal one] image

Commands used to train Goyal, python main.py data/Goyal --workspace model/trial_may -O --iters 60000 --asr_model ave --preload 2 For test, python main.py data/Goyal --workspace model/trial_may -O --test --asr_model ave --portrait --aud ./demo/test.wav

I wonder why are we loading the 19th epoch every time, why there are no epoch training for my custom data so that I can load that?

image

For more information, please refer https://github.com/ZiqiaoPeng/SyncTalk/issues/164

ZiqiaoPeng commented 4 months ago

modify '--workspace (path to your model)'

HardikJain02 commented 4 months ago

my model hasn't been generated. after loading training data it says Killed. [Refer issue 164

My machine specifications image