tijiang13 / InstantAvatar

InstantAvatar: Learning Avatars from Monocular Video in 60 Seconds (CVPR 2023)
376 stars 32 forks source link

Error when increasing batch size #55

Closed herybala closed 1 year ago

herybala commented 1 year ago

I saw we get error when increasing batch size in dataloader

tijiang13 commented 1 year ago

Hi Herybala,

Yes, in the code, we assume the batch size to be 1 (which means we train with 1 frame at a time). However, you can increase the number of pixels you want to sample like in here (4 32 32 = 4K by default)

Best, Tianjian

herybala commented 1 year ago

Thanks for advice

Le mar. 12 sept. 2023 à 13:26, Tianjian Jiang @.***> a écrit :

Hi Herybala,

Yes, in the code, we assume the batch size to be 1 (which means we train with 1 frame at a time). However, you can increase the number of pixels you want to sample like in here https://github.com/tijiang13/InstantAvatar/blob/master/confs/sampler/patch.yaml#L3 (4K by default)

Best, Tianjian

— Reply to this email directly, view it on GitHub https://github.com/tijiang13/InstantAvatar/issues/55#issuecomment-1715544154, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJVPXZFIEQ4SF67SPRCJETLX2BBHZANCNFSM6AAAAAA4UTEOBM . You are receiving this because you authored the thread.Message ID: @.***>

-- Hery Velonjatovo

herybala commented 1 year ago

Sorry wrong manipulation, i found the problem Hello, Also when setting sampler to edge i got the following error:

hydra.errors.ConfigCompositionException: Could not override 'sampler.dilate'. To append to your config use +sampler.dilate=8

tijiang13 commented 1 year ago

Hi Herybala,

From the error it appears that you are trying to pass parameters that EdgeSampler doesn't recognise to it (dilate in this case). If you want to switch sampler you can either:

  1. change "patch" to "edge" here
  2. or specify in the command line like "python train.py sampler=edge"

Best, Tianjian