Allow passing fused point cloud in the training

[ApdowJN]

Summary

This merge request adds support for fused point cloud initialization .

---

Features Added

• Fused Point Cloud Initialization
  Introduced a new initialization method, fused_point_cloud, enabling initialization of 3D Gaussians from a point cloud (.ply file).

• cuSFM Output Adaptation
  Modified the line parsing logic in function read_colmap_extrinsics_text to handle cuSFM’s images.txt format, where some images contain only the line
  IMAGE_ID, QW, QX, QY, QZ, TX, TY, TZ, CAMERA_ID, NAME
  and do not include the POINTS2D[] line (which normally contains (X, Y, POINT3D_ID)).

  # Original
  lines = (line for line in lines if line and not line.startswith("#"))
  
  # Updated
  lines = (line for line in lines if line == "" or not line.startswith("#"))

---

Config Files Using Accumulated Point Cloud Initialization

• Added configs/apps/cusfm_3dgut_mcmc.yaml — configuration for accumulated point cloud–based training.

• Added configs/initialization/accumulated_point_cloud.yaml — defines parameters for point cloud initialization:

  • observation_scale_factor: 0.01
  • use_observation_points: true
  • accumulated_point_cloud_path: configurable path to the input .ply file

Commands to train 3DGRUT using a given point cloud for initialization

  python train.py \
    --config-name apps/cusfm_3dgut_mcmc.yaml \
    path=<path_to_cuSFM> \
    out_dir=<path_to_out> \  
    initialization.accumulated_point_cloud_path=<path_to_nvblox_dir>/nvblox_mesh.ply \
    experiment_name=3dgut_mcmc \
    export_usdz.enabled=true \
    export_usdz.apply_normalizing_transform=true

[wilsonCernWq]

Just curious, what does this line fix?

[ApdowJN]

Hi @wilsonCernWq, Thank you for your review.
This update handles cases in images.txt where no 2D points are associated with an image.
In such cases, some entries contain only:
IMAGE_ID, QW, QX, QY, QZ, TX, TY, TZ, CAMERA_ID, NAME
and do not include the usual POINTS2D[] line (which normally lists (X, Y, POINT3D_ID)).
A line was added to properly handle these cases during file reading.

[wilsonCernWq]

In this case, would we remove line 517 (the comment) and add actual comments to it to explain a bit? Sorry for the trouble! Thank you!

[ApdowJN]

Sure, will do that. Thanks.

[ApdowJN]

Hi @wilsonCernWq I updated the following:

1. Replaced “accumulated” with “fused” to avoid ambiguity.
2. Added comments to clarify line 518.

Could you please take a look? Thank you! If everything looks good, could you help me merge it?

[wilsonCernWq]

Looks good to me, thanks for accommodating the change request