search

chungyiweng / humannerf

HumanNeRF turns a monocular video of moving people into a 360 free-viewpoint video.
MIT License
786 stars 86 forks source link

Issue about results on custom monocular video #41

Open Auggst opened 1 year ago

Auggst commented 1 year ago

I try to train a model on the people_snapshot dataset, but the results is so bad, as following: image where the odd-numbered columns are the model output results, and the even-numbered columns are gt

This is the link of result, check point and dataset i processed using your method: https://1drv.ms/u/s!AsI-fRFqtwxTnEF1FnO9srkEeRdM?e=6sh1yU

Is there something wrong with my processing of the dataset?

Dipankar1997161 commented 1 year ago

@Auggst,

Hello, Could you tell me, from where you got the smpl parameters and the mask images for custom training.

Also, the camera parameters, did you take the people snapshot one or just set extrinsic to np.eye(4) Would be helpful for me