Repository navigation
v5.18.0 — VLM Check: AI Visual Anomaly Detection
VLM Check — Claude Vision Anomaly Detection (v5.18.0, build 61)
AI-powered visual anomaly detection for astrophotography sessions. Generates mosaic wallpapers from cached frames and uses Claude Vision (Opus with extended thinking) to detect optical defects that metrics alone can't catch.
New Features
- VLM Check toolbar button — generates tiled mosaics grouped by target/filter/setup
- Deviation Map — per-pixel median comparison with heat map visualization (toggle via waveform button). Makes ice shadows, dew, clouds instantly obvious
- Click-to-Mark tiles — single-click any tile to mark/unmark the corresponding frame in the file list (blue overlay)
- Chronological sequence analysis — detects transient defects that appear/persist/worsen over time
- Anomaly types: ice crystals, dew, clouds, obstructions, focus shifts, light leaks
- Supabase edge function — 10 free VLM checks/day, fallback to user's own Claude API key
- Clean overlay — red rectangle border + type/confidence label (no fill, tiles stay visible)
- Floating validation window — group tabs, zoomable mosaic, anomaly list with jump-to-frame
Technical
- Tiles: 480×360px, center-cropped 80%, annotated with frame number, time, moon distance, twilight
- Extended thinking (10K token budget) for systematic tile-by-tile analysis
- Deviation map: pixel-by-pixel median, gamma-enhanced, hot palette visualization