search

Songluchuan / AdaSR-TalkingHead

ICASSP2024: Adaptive Super Resolution For One-Shot Talking-Head Generation
168 stars 13 forks source link

What is the loss_weights['reconstruction'] during training? #3

Open Guanyuansheng opened 6 months ago

Guanyuansheng commented 6 months ago

This weight seems to be missing from the config file

Songluchuan commented 6 months ago

I follow the face2vid to finish my loss function, please refer to the https://github.com/zhanglonghao1992/One-Shot_Free-View_Neural_Talking_Head_Synthesis for details.

Guanyuansheng commented 6 months ago

I follow the face2vid to finish my loss function, please refer to the https://github.com/zhanglonghao1992/One-Shot_Free-View_Neural_Talking_Head_Synthesis for details.

As far as I know, L1 loss is not used in the facevid2vid. I would like to ask what benefits will be brought by adding L1 loss, and what is the weight of L1 loss?