search

yhw-yhw / TalkSHOW

This is the official repository for TalkSHOW: Generating Holistic 3D Human Motion from Speech [CVPR2023].
301 stars 26 forks source link

Different Results with Paper #4

Open JeremyCJM opened 1 year ago

JeremyCJM commented 1 year ago

I got very different metric values for the face using your provided checkpoint, compared with the values on paper.

On paper: image

I tested: image

zjchu commented 1 year ago

image This is the result of retraining the smplx_face model, which provides very different metric values compared with the values on paper.

zjchu commented 1 year ago

I got very different metric values for the face using your provided checkpoint, compared with the values on paper.

On paper: image

I tested: image

I think when this code ( python data_utils/dataset_preprocess.py) is executed, it will split train dataset, valid dataset and test dataset randomly. Some of the test data you use may have been used to train the provided checkpoint, but the LVD values are about a tenth of the ones in the paper.