---

name: visualization-and-results description: "Interpret UniAD result artifacts and build visualization/log-analysis commands." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

visualization-and-results

Use this sub-skill when the user already has a UniAD result artifact and needs to inspect it, render it, or build a visualization command.

Do this first

• Inspect the pickle with scripts/inspect_results_pickle.py if the structure is unknown or keys look missing.
• Build the visualization command with scripts/build_visualization_command.py.
• Treat tools/test.py --out as the normal source of the visualization pickle.
• Treat --show-dir as a separate test-time image sink, not the visualizer input.

Route away when needed

• Generating a fresh result pickle via evaluation -> training-evaluation
• Config or model edits, or missing task heads -> config-and-model-architecture
• Dataset layout, info PKLs, or motion-anchor problems -> data-preparation

What this sub-skill understands

• tools/test.py --out result pickle behavior
• tools/test.py --show-dir as a separate review path
• visualization runner args: predroot, out_folder, demo_video, project_to_cam
• UniAD forward-test result keys for tracking, motion, map, planning, and occupancy
• log-analysis as a secondary check when the pickle structure is fine but the render looks wrong

Expected outputs

• per-sample JPG frames in the output folder
• optional combined BEV/camera JPGs
• a final AVI video
• a compact pickle key summary for troubleshooting

References

• references/visualization.md
• references/results-format.md
• references/troubleshooting.md

Scripts

• scripts/build_visualization_command.py
• scripts/inspect_results_pickle.py