Implementation of sparse depth loss breaks the code

[Ahmad-Jarrar]

The provided script for synthetic data scripts/synthetic_nerf_run.sh worked well until commit 37a93672d5464eb0781d910a156075b9017b1c32. The issue has started with the implementation of sparse depth loss 12dd57be09488c390eaeff2aa01e4d7a082d8573.

============ start lego ============
Output folder: output/synthetic/baseline/lego
Read dataset in NeRF format.
res_downscale=0.0
res_width=0
Traceback (most recent call last):
File "/local/home/svraster/train.py", line 584, in
training(args)
File "/local/home/svraster/train.py", line 48, in training
data_pack = DataPack(cfg.data, cfg.model.white_background)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/local/home/svraster/src/dataloader/data_pack.py", line 61, in init
self._cameras['train'][dataset_downscale] = CameraList(
^^^^^^^^^^^
File "/local/home/svraster/src/dataloader/data_pack.py", line 117, in init
self.camera_list = [
^
File "/local/home/svraster/src/dataloader/data_pack.py", line 118, in
instantiate_a_camera(cam_info, cfg_data, dataset_downscale)
File "/local/home/svraster/src/dataloader/data_pack.py", line 187, in instantiate_a_camera
return Camera(w2c=cam_info.w2c,
^^^^^^^^^^^^^^^^^^^^^^^^
File "/local/home/svraster/src/cameras.py", line 166, in init
self.sparse_uv = torch.tensor(sparse_uv, dtype=torch.float32, device="cpu")
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: must be real number, not NoneType

[Atehortuajf]

I have the same stack trace when trying to train my own dataset generated from the InstantNGP colmap2nerf.py script. Any help would be appreciated.