search

xxlong0 / SparseNeuS

SparseNeuS: Fast Generalizable Neural Surface Reconstruction from Sparse views
MIT License
319 stars 16 forks source link

I want to try my dataset and hope you give some advice? #19

Open MNILjj opened 1 year ago

MNILjj commented 1 year ago

Hello, I want to try train my dataset,but I run into some problems,hope you can give me some advice？Thanks.

1)The camera internal and external parameters used in my data are not uniform, so I modified some of the code in "dtu_general.py", I am more confused about the role of the "pair.txt" file in it, whether each Does the scene need one?

2）The resolution of the image is not 640x512, not a multiple of 32. Does running the program have any effect?

3）I selected 9 sets of data in the dtu data set to remove the background for training. The results obtained were very bad. Do you need to increase the amount of data or why? 2022-09-05 19-15-12 的屏幕截图 Screenshot 2022-09-05 19:12:45

flamehaze1115 commented 1 year ago

1), the "pair.txt" file in DTU records the pairs of one reference image and its source images, which makes sure the input images have enough overlaps. You should make sure the input images of your dataset do so. 2) The resolution won't be a problem, just make sure its resolution is OK for the 2D feature network. 3) You can first try to fit the network on one pair of images and see the results to check for any problems.

MNILjj commented 1 year ago

Thanks! I chose to train only one as you said. But the result is not very good. I would like to ask if "meshes_train_bg" and "meshes_val_bg" are 3D models of objects? 13d40fbb1ee78e90d1a4a94efbca915 屏幕截图 2022-09-06 105839

flamehaze1115 commented 1 year ago

You can set the weight of sparse loss to 0 to see the results, since this term will penalize all the sdf values and make the training difficult.

Yes. "meshes_train_bg" and "meshes_val_bg" are 3D models of objects.