Using Pre-processed Data

For advanced users who have already processed their images to calculate camera poses, Splattr offers a "Pre-processed" project type. This workflow allows you to skip our data processing stage entirely, saving both (a lot of) time and credits.

The "NeRF Studio" Format

Please see https://docs.nerf.studio/quickstart/data_conventions.html

Splattr is compatible with the standard dataset format used by many popular tools, including Nerfstudio, Kiri Engine, Polycam, Luma AI, and raw COLMAP outputs.

The core requirement is a specific folder structure containing your images and a transforms.json file.

Required File Structure

To be valid, your project must be compressed into a single .zip file with the following internal structure:

my_project.zip
├── images/
│ ├── frame_0001.jpg
│ ├── frame_0002.jpg
│ └── ...
└── transforms.json

• transforms.json: This file must be at the root of the ZIP archive. It contains the camera intrinsics and the calculated camera-to-world transformation matrix for every image in your scene.
• images/ folder: This folder (which can also be named image/) contains all the source images referenced within transforms.json. The file paths inside transforms.json must correctly point to the images within this folder (e.g., "file_path": "images/frame_0001.jpg").

How to Create a Pre-processed Project

1. Prepare Your Data: Ensure your processed data matches the required file structure and compress it into a single .zip file.
2. Select Project Type: In the Splattr "New Splat" flow, choose the "Processed Data" option. This will configure the job to skip the data processing stage.
3. Upload & Configure: Upload your .zip file. Our system will validate it on the client side to give you an instant price quote. You can then configure your Training settings (Training Method, Iterations, Compute Tier) as usual.
4. Start Your Job: The final credit cost will only be for the training stage, making it significantly cheaper than a standard project.

Benefits of Using Pre-processed Data

• Save Credits: You are not charged for the data processing or feature matching stages.
• Use External Tools: It allows you to use specialized tools like RealityCapture, Metashape, or your own custom COLMAP scripts to generate camera poses, giving you maximum control over the pre-processing pipeline.
• Interoperability: Easily move projects between Splattr and other platforms that support the NeRF Studio format.

Troubleshooting

• "transforms.json not found": This error means the transforms.json file is not at the root of your ZIP archive. Re-compress your files, ensuring it is not inside any sub-folders.
• "Image not found": This error means a file path listed in transforms.json (e.g., images/frame_0001.jpg) does not exist at that location within your ZIP file. Please verify that all image paths are correct and all images are included.