There are many holes in the mesh of the dtu dataset.

bennyguo / instant-nsr-pl

Neural Surface reconstruction based on Instant-NGP. Efficient and customizable boilerplate for your research projects. Train NeuS in 10min!

MIT License

856 stars 84 forks source link

There are many holes in the mesh of the dtu dataset. #79

Open ZZy129326999 opened 1 year ago

ZZy129326999 commented 1 year ago

Thank for your excellent work, I wonder why there are many holes in the mesh of the dtu dataset? I noticed that the inv_s is indeed large enough during training. Thank you very much!!

YZsZY commented 1 year ago

I have also encountered this problem, if you have a solution, please @me haha https://github.com/bennyguo/instant-nsr-pl/issues/59

bennyguo commented 1 year ago

Actually I have this kind of problem too. I suspect it's because of the high-frequency nature of the hash grid. You may try increasing the update_steps in the xyz_encoding_config and see if you get any luck :)