Side experiments: models and scripts that support image handling, tagging, and workflow—not production claims.
Overview
This page collects exploratory work: training small classifiers on render passes, testing off-the-shelf segmentation for masking, and notebook workflows that turn folders of frames into charts (histograms, brightness drift between shots). The point is learning what helps a solo pipeline and what is not worth the maintenance.
Directions
- Labelling subsets of project renders to compare lighting setups across scenes.
- Trying automatic captions or tags for archive search—not for client-facing copy.
- Measuring inference cost vs. time saved before committing to a step in production.
When a notebook graduates to a script, it moves next to Pilgrim Image Converter and other tooling; anything fragile stays here as notes.