About Fitting on New Datasets

YuelangX / Multiview-3DMM-Fitting

Fitting 3DMM models to multiview (monocular) video data.

MIT License

59 stars 10 forks source link

About Fitting on New Datasets #1

Closed jeb0813 closed 1 month ago

jeb0813 commented 3 months ago

Hi Yuelang.

I followed the instructions and obtained identical results on NeRSemble, but when I tried it on MEAD, the loss starts at around 750 and never converg.

The preprocess on these two datasets is same, expect the camera params, I got them from colmap (but the cam params in NeRSemble is also from colmap).

And I have checked the lmds from step 1, seems alright. However, the fitting cannot work. Do you have any idea on that?

Thanks.

YuelangX commented 3 months ago

What does the visualization result look like? You can try adjusting the pose and scale initialized in BFMModule.py.

jeb0813 commented 3 months ago

@YuelangX thanks for quickly response. The visualization seems fine on both datasets.

In your approach, is the BFM model aligned to the world coordinate system when initialized? I think this has something to do with the camera's external parameters.