search

snap-research / articulated-animation

Code for Motion Representations for Articulated Animation paper
https://snap-research.github.io/articulated-animation/
Other
1.21k stars 348 forks source link

output of 512x512 for vox model #18

Open akashdexati opened 3 years ago

akashdexati commented 3 years ago

Hi @AliaksandrSiarohin Is it possible to attain output of 512x512, by retraining using vox256 checkpoint file ? Please suggest what can be done for 512x512 output.

Thanks

AliaksandrSiarohin commented 3 years ago

No, just change the files to 512x512.

leiluoray1 commented 2 years ago

Hi @AliaksandrSiarohin. Thanks for the wonderful research. What advices would you give for training the model for 512x512 data sets? use training data of 512x512? any modifications to the original training config? anything else?

AliaksandrSiarohin commented 2 years ago

512^2 usually have worse regions than 256^2. So it is probably better to first train on 256^2 and then finetune on 512^2.

leiluoray1 commented 2 years ago

@AliaksandrSiarohin Thanks for the quick reply. Wouldn't the layers have different shapes so the per-trained weights won't be compatible? Could you please elaborate how I would do that? Thanks.