5.4: Scene-Type Detection And Training Profiles

[kmeirlaen]

Included survey bullets:

• Auto-detect outdoor, indoor, or single-object scenes.
• Adjust training parameters accordingly.
• Derive max splat count from scene scale and initial point cloud rather than hardcoding.
• Save and select processing profiles for different cameras/drones.
• Settings import/export may already exist but is hidden.

Suggested solution:

Split this into two layers: make profiles visible first, then add recommendations or auto-detection. Avoid pretending automatic scene classification is smarter than it is; present it as suggested settings the user can accept or modify.

Actionable scope:

• Surface existing config import/export as named profiles.
• Add profile presets for common capture types.
• Estimate scene scale and initial point-cloud density.
• Suggest max splat count and related parameters.
• Allow user override.

Acceptance signal:

• A user can save, select, and reuse named settings profiles.
• LichtFeld can recommend starting parameters based on scene cues without forcing them.

Dependencies:

• Existing config import/export.
• Training parameter UI.
• Dataset analysis.