The depth maps exported by Nerfstudio are usually color images. Is it possible to export the raw depth images?

[Lizhinwafu]

The depth maps exported by Nerfstudio are usually false color images. Is it possible to export the raw depth images?

[Tavish9]

The depth maps exported by Nerfstudio are usually false color images. Is it possible to export the raw depth images?

Hi, @Lizhinwafu, Nerfstudio applies a colormap to the image after rendering. For more details, please refer to utils/colormaps.py. If you would like to save the raw depth as a .npy file, you will need to manually modify the corresponding code.

[Lizhinwafu]

I want to save the raw depth map as **.jpg. Which code need to modify?

[Tavish9]

I want to save the raw depth map as **.jpg. Which code need to modify?

By “raw depth image,” are you referring to saving the depth map in its unprocessed form? If you’re looking to save the depth map as a grayscale image, you can simply set ColormapOptions.colormap = "gray". On the other hand, if you’re aiming to save the raw depth data as a .npy file, you’ll need to modify the code at this block.

[Lizhinwafu]

Thanks. We can directly get RGB and depth using a depth camera. My goal is to generate the same depth as the camera. Pseudo-color images can only be visualized. I want to combine the rendered depth map and RGB to generate the point cloud myself.

[Tavish9]

Thanks. We can directly get RGB and depth using a depth camera. My goal is to generate the same depth as the camera. Pseudo-color images can only be visualized. I want to combine the rendered depth map and RGB to generate the point cloud myself.

What is the data type and shape of your depth got by the camera? The depth generated by NeRF is of type float and has a shape of (h,w,1). After applying a colormap to the raw depth, the shape changes to (h,w,3).

[Lizhinwafu]

I rendered RGB and depth, and generated point clouds from multiple views using the intrinsic and extrinsic parameters estimated by COLMAP. Why do the point clouds generated from multiple views not overlap with each other?

[Lizhinwafu]

When using Nerfstudio for data preprocessing, it generates a transform.json file, and I have also obtained the RGB and depth images for each view, which I used to generate point clouds for each view. How can I use the transform.json file to register the point clouds from multiple views into a single unified point cloud?

[Tavish9]

When using Nerfstudio for data preprocessing, it generates a transform.json file, and I have also obtained the RGB and depth images for each view, which I used to generate point clouds for each view. How can I use the transform.json file to register the point clouds from multiple views into a single unified point cloud?

Hi, sorry for late reply.

If you have already train your nerf model, you can generate the pointcloud using ns-export pointcloud --load-config path/to/config.yml --output-dir exports/pcd/ --num-points 100000 --remove-outliers True --normal-method open3d --save-world-frame False.

This command would generate only one unified pointcloud.