One-Shot_Free-View_Neural_Talking_Head_Synthesis icon indicating copy to clipboard operation
One-Shot_Free-View_Neural_Talking_Head_Synthesis copied to clipboard

Published 20 hours ago • verified publisher icon zhanglonghao1992

Is the pre-trained model configured for the SPADE mode

Open qianyuling-wl opened this issue 1 year ago • 0 comments

I apologize for the intrusion, but I wanted to inquire about the pre-trained model you provided. Is it configured for the SPADE mode? I'm encountering some issues while training in SPADE mode. The only change I made was to slightly decrease the 'num_repeats' parameter in the training settings. However, the model I trained is not producing the expected results; the mouth shapes don't seem to be driving correctly. Is there a difference between setting num_repeat to 75 with 200 epochs and setting num_repeat to 10 with 1500 epochs in the training configuration? I was wondering if you could possibly help me troubleshoot this issue. Attached are the results generated by the model along with the loss curve for your reference. Thank you for your time and assistance. loss

https://github.com/zhanglonghao1992/One-Shot_Free-View_Neural_Talking_Head_Synthesis/assets/82922019/5e9a3515-223b-4212-815c-80fd6a758810

https://github.com/zhanglonghao1992/One-Shot_Free-View_Neural_Talking_Head_Synthesis/assets/82922019/9b5f3325-1859-4a52-bb48-2892bd0ab625

qianyuling-wl avatar Nov 12 '23 02:11 qianyuling-wl