---

name: sunrgbd description: "Navigate the beta SUN RGB-D preparation, one-hot training, testing, visualization, and 3D AP evaluation code included with Frustum PointNets." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

SUN RGB-D

Use this route only for the repository's supplementary SUN RGB-D pipeline. The source labels it a beta release and does not provide a turnkey raw-data setup: it expects the external SUNRGBD V1 dataset/toolbox, MATLAB extraction, a manually organized training tree, generated pickle assets, and detector boxes.

Route

1. Read data and workflows and run python scripts/check_sunrgbd_inputs.py --help before preprocessing.
2. Treat MATLAB/toolbox extraction as an external prerequisite. Do not copy the vendored toolbox binaries into an operating environment.
3. Generate/validate the appropriate training or detector-frustum pickle.
4. Route TensorFlow setup to ../runtime-and-custom-ops/SKILL.md and shared training decisions to ../training/SKILL.md.
5. Use the one-hot test path with --dump_result, then read evaluation before computing 3D AP.

Visualization is optional, GUI-dependent, and not a verification gate. No raw SUN RGB-D preparation, long training, or AP benchmark was executed while this skill was created. Read troubleshooting for missing external assets, Python-2 pickle issues, class/schema mismatch, and headless Mayavi failures.