search

autonomousvision / monosdf

[NeurIPS'22] MonoSDF: Exploring Monocular Geometric Cues for Neural Implicit Surface Reconstruction
MIT License
565 stars 52 forks source link

Problems occurred while training my own dataset #87

Open wujinhu1999 opened 1 year ago

wujinhu1999 commented 1 year ago

Output during training： merge_410_6 image The configuration I use is dtu_mlp_fullres_allviews.conf, which is a dataset around the brain without a mask image. I did not run scannet_to_monosdf.py, I changed img_res in dtu_mlp_fullres_allviews.conf to [384, 384].

niujinshuchong commented 1 year ago

How many images do you use for training?

wujinhu1999 commented 1 year ago

How many images do you use for training?

hi,I use 68 images for training.

niujinshuchong commented 1 year ago

Could you first check if your camera poses are correct. Other than that, maybe you could try to use a smaller sphere at initialization.

wujinhu1999 commented 1 year ago

Could you first check if your camera poses are correct. Other than that, maybe you could try to use a smaller sphere at initialization. I would like to ask how to use a smaller sphere at initialization and why do that.

ansj11 commented 7 months ago

I also has the same problem, did you solve the problem?