Kai Zhang
hmmm, I also found that finetuning the pretrained weight on customized RGBD dataset fails to converge, leading to all-black predictions for some reason. In my case, the model even cannot...
I believe there is also a bug in the compute_l2_distance function. One correct implementation can be: ```python def compute_l2_distance(x, y): N, C, H, W = x.size() x_vec = x.view(N, C,...
This script from my latest IRON work might be helpful: https://github.com/Kai-46/IRON/blob/main/models/export_materials.py
We assume the inputs do not include object segmentation masks; hence the mask loss in IDR is not applicable.
Once you render the camera path, you can use the following script to create a video. ```python import imageio import os render_dir = '/path/to/rendered/camera/path' frames = [] fps = 10...
Sorry, I'm not sure if I can find the scripts. But training a opacity field to a unit sphere seems very straightforward? You can simply train the MLP using a...
Did you set up the lighting properly?
What is your camera coordinate convention?
Please check this git issue: https://github.com/Kai-46/ARF-svox2/issues/11
Note at least 12GB GPU memory is needed for the default settings. You can try decreasing the rendered patch size if you have less memory.