---

name: visualization-and-analysis description: "Route Det3D LiDAR, BEV, KITTI, prediction, training-log, FLOPs, and headless-visualization analysis tasks." metadata: disco-role: operating disable-model-invocation: true license: Apache 2.0

Visualization and Analysis

Use this route for prediction/result rendering, BEV/LiDAR/KITTI visualization, training-log summaries, learning curves, and model-complexity interpretation. Use scripts/summarize_log.py for deterministic headless log summaries.

Workflow

1. Identify artifact type: JSON log, pickle/result objects, KITTI text, point-cloud/boxes, image/calibration, or model/config.
2. Validate coordinate frames, box convention, classes, and score threshold before rendering.
3. Prefer saved images/text summaries on remote or headless hosts.
4. Read visualization-workflows.md for optional VTK/Open3D/Matplotlib paths and log-analysis.md for scalar aggregation.
5. Follow troubleshooting.md for display, dependency, malformed-log, or empty-result failures.

FLOPs and parameter counts depend on the exact model/config/input shape and may require the same compiled dependencies as model construction. A plot or summary does not prove detector correctness; route metric/evaluator concerns to training-and-evaluation.