skipping get sparse

[hey0chen]

Hi,
Thanks for your great work! I noticed that it's unnecessary to use colmap to get sparse if I already had the right poses, such as when I use spinnerf_datase. So I want to skip this step and use the dataset directly, but the file names are not matched with fellow steps.
Could you solve this problem? I think it is important to use the poses provided by dataset, instead of using colmap to get poses that may not accurate.
Thanks for your time and consideration.

[cuteyyt]

Thanks for your interest in our project!

We reconstruct camera parameters for the spinnerf dataset because we are unsure whether mixing the views before and after deletion altogether is appropriate. If you check the camera files provided by the spinnerf dataset, you can find they have 100 views (40 deleted and 60 original) as one scene. I guess the spinnerf dataset gets its camera parameters through COLMAP in an LLFF way with the 'images' folder directly. Due to this concern, we reconstruct camera parameters from the 60 original views ourselves.

Regarding your request to use the camera parameters provided by the spinnerf dataset directly, I am willing to help, but I am a little busy these days. I will try to add this feature in early August, or if you have already started to do this, I would be happy to discuss it with you and would appreciate it if you could share your progress.

[cuteyyt]

Due to my schedule, I may be unable to readjust the code of the project recently. Sorry for this, and I can provide some hints here to skip the sparse folder. The key point is to ensure that the camera parameters provided by the dataset itself match the image file name exactly. As the datasets used in this project run the COLMAP with the 'images' folder, and we may wish to run on a down-sampled image size such as 'images_4', errors may occur if the above two folders use different image file names. A possible solution is to make a name mapping between the filenames appearing in camera parameters and the image folder, such as:

path_mapping = {Path(ori_img_path).name: Path(img_path).name for ori_img_path, img_path in zip(ori_img_paths, img_paths)}

Here ori_img_path is the path used in camera parameters, and img_path is the path used for the down-sampled images. Then you can use this dict to ensure the camera parameter matches the correct image.

I hope this can help you, and I will try to improve this project with your suggestion as early as possible.