Unable to Reproduce Results with Provided Configurations on H3DS

[Darshangohel2610]

Hello @KeyuWu-CS,

Thank you for your excellent work and for open-sourcing this project. I have been trying to reproduce the results as outlined in your paper but encountered some difficulties. Specifically, I used the given configurations from the H3DS example config file, but the outputs do not match the quality or accuracy shown in the paper.

What I Tried:
I have followed the steps in the documentation precisely.
Used the provided configuration files(all given examples) with/without modifications.

Could you kindly provide the exact configuration file used to generate the results in the paper? Any insights or tips to help reproduce the results would be greatly appreciated.

Thanks in advance for your time and support!
Thank you

[KeyuWu-CS]

Sorry for later response since it's busy recently. For H3DS, there are two important solutions to improve quality

1. H3DS have few images (32, obvious not enough) and the lighting is constantly changing. So if you use COLMAP to get the camera pose, the result will be very bad if I remember correctly. It is recommended that you use the more accurate camera pose they provide (this may take some time to try how to scale and transform the coordinate system).
2. The higher the quality of hair segmentation, the better. H3DS has many inaccuracies in hair segmentation. It is recommended that you use SAM to segment them.
   But also remember check if the head fitting step get a reasonable results. Good luck.