Skip to content

Conversation

@Asnly1
Copy link

@Asnly1 Asnly1 commented Nov 17, 2025

Here is a summary of the new features:

1. Matrix-Driven Camera Mode

  • This new mode allows rendering a dataset (train/test) directly from a transforms.json file with same format in your original implementation.
  • It automatically restores camera poses and intrinsics, making it possible to re-render scenes from existing camera trajectories.

2. Multi-Channel Data Export

  • The rendering pipeline has been re-engineered to simultaneously export multiple data channels.
  • Users can now enable checkboxes in the UI to export:
    • RGB (.png)
    • Instance Mask (.png)
    • Depth (.png and .exr)
    • Normals (.png and .exr)

3. Expanded Camera Paths for COS Mode

  • The "Circle on Sphere" (COS) mode has been extended with more camera path options:
    • Spherical Spiral Trajectory: Smoothly covers both the upper and lower hemispheres.
    • Multi-Level Horizontal Orbits: Allows capturing images at several distinct height levels, simulating common drone capture patterns.

4. Workflow & Quality-of-Life Improvements

  • Test Set Consistency: The transforms_test.json generated by SOF/TTC/COS modes can now reuse an existing matrix file, ensuring consistent test sets across different training data acquisitions.
  • Enhanced Logging: Log files now include the full camera intrinsic matrix for easier debugging.
  • Optional Dataset Compression: Users can choose to automatically compress the dataset into a .zip file and delete the raw image folders upon completion.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant