---

name: data-preparation description: "Guides Nerfstudio dataset conversion, transforms.json validation, camera conventions, device capture modes, and dataparser selection." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Data Preparation

Use this route before training when the task is about custom captures, camera poses, transforms.json, ns-process-data, or choosing a dataparser.

What this route covers

• Converting images, videos, Polycam, Record3D, Metashape, RealityCapture, ODM, and Project Aria style captures into the Nerfstudio dataset layout.
• Validating the Nerfstudio transforms*.json schema, image/depth/mask paths, split lists, and camera conventions.
• Distinguishing COLMAP/FFmpeg/hloc/external-tool requirements from package-level Python issues.
• Choosing between nerfstudio-data, blender-data, minimal-parser, colmap, and other built-in dataparser names.

What this route excludes

• Training method choice and config overrides: use training-and-configs after the dataset validates.
• Dataset download commands: use cli-workflows first, because downloads need network and disk checks.
• Custom dataparser packaging: use api-extension.

Read/run these bundled files

• references/data-formats.md for transforms*.json, coordinate, depth, mask, and split fields.
• references/workflows.md for conversion recipes and dataparser routing.
• references/troubleshooting.md for COLMAP, FFmpeg, path, and schema failures.
• scripts/validate_nerfstudio_dataset.py to check a dataset directory or transforms*.json without modifying data.

Safe workflow

1. Identify the capture type and external prerequisites.
2. If using images/video, confirm COLMAP and FFmpeg unless an existing sparse model makes --skip-colmap valid.
3. Run the appropriate ns-process-data command into a new output directory.
4. Validate the output directory with the bundled validator.
5. Pass the dataset directory to ns-train through training-and-configs.

Key reminders

• Nerfstudio uses OpenGL/Blender-style camera coordinates; COLMAP/OpenCV axes differ and are converted by the processing/dataparser code.
• Image paths in transforms*.json are relative to the dataset directory unless absolute paths are intentionally used.
• Depth images are specified per frame with depth_file_path and are interpreted in millimeters by default.
• Masks must be one-channel black/white images, same resolution as RGB, and must be present for all frames if used.