Understanding Input Types

Splattr supports several types of input data, each with its own workflow.

Photos

• Format: A .zip archive containing a collection of images (.jpg, .png).
• Workflow: This is a "Standard" project. Splattr will run the full data processing pipeline to analyze your images, calculate camera positions, and then train the splat.
• Best for: Unordered photo sets taken with a DSLR or smartphone.

Videos

• Format: A video file (.mp4, .mov, .avi, .webm).
• Workflow: This is also a "Standard" project. Splattr will first extract frames from your video based on your "Frame Sampling Rate" setting, then proceed with the full data processing and training pipeline.
• Best for: Capturing scenes with smooth, continuous camera motion.

Pre-aligned Data

For Advanced Users

This workflow is for users who have already processed their images with other software and have calculated camera poses.

• Format: A .zip archive containing a specific file that describes the camera positions.
• Workflow: This workflow skips the expensive "Data Processing" stage, saving you significant time and credits. Splattr will read the provided camera data and proceed directly to the "Training" stage.
• Supported Formats:
  • transforms.json: The standard format used by Nerfstudio and other popular tools.
  • Metashape: A .zip containing a cameras.xml (you can name it whatever you want) file exported from Agisoft Metashape.
  • RealityScan / RealityCapture: A .zip containing a csv file (named however you want) exported from the application.