point_based_clothing icon indicating copy to clipboard operation
point_based_clothing copied to clipboard

Published 20 hours ago • verified publisher icon saic-vul

Custom Data

Open sidk99 opened this issue 2 years ago • 3 comments

Trying to run inference on custom data. However input seems to require additional segmentation data, that doesn't seem to be specified. What preprocessing steps are required?

sidk99 avatar Jan 10 '22 20:01 sidk99

Hello @sidk99,

Thank you for your interest in our work!

Sorry for the late reply, here are the instructions that I added to README on how to get segmentations in the required format. After this procedure, the segmentation mask is loaded in load_segm() and its result (which is simply a binary mask of lower+upper clothing) is used in a dataloader for model training or inference.

Also, make sure that you have the SMPL-parameters of a person on your image or video frame.

Please let me know if something else is needed to run the pipeline on custom data.

izakharkin avatar Jan 19 '22 20:01 izakharkin

Hello @izakharkin I have also tried to run inference on custom data, but I'm not able to get the SMPL-parameters. Could you please provide a repository or some instructions on how to obtain SMPL information?

tabatabaeifateme avatar Jan 24 '22 09:01 tabatabaeifateme

Hello @tabatabaeifateme,

Sure, there is actually much open-source code that allows getting SMPL-parameters from an image or video. Probably one of the easiest ways is to run FrankMocap, the parameters are being saved to a separate .pkl in this line. Please make sure that you predict SMPL parameters, not SMPL-X parameters (their options allows you to choose).

Please also make sure that you convert the predicted SMPL-parameters to our format.

izakharkin avatar Jan 24 '22 15:01 izakharkin