visual-attention-auditlisted

Measure a design's visual clutter and where the eye lands, with numbers instead of opinion. Computes the two published Rosenholtz clutter measures (Feature Congestion and Subband Entropy) plus colour mass on a blurred frame, and ranks a frame against its neighbours in a grid (search results, a shelf, an icon page, a competitor set). Works on any image: UI screenshot, landing page, banner, product card, packaging. Triggers: "is this design too busy", "measure visual clutter", "which element draws the eye first", "does my design stand out among competitors", "squint test", "visual noise score", "attention audit", "why does this screen feel cluttered", "compare my thumbnail against the competition". NOT an eye-tracking replacement and NOT a saliency model trained on gaze data; see "What this is not" below.
AlexeyGOblov/visual-attention-audit · ★ 0 · AI & Automation · score 72

Install: claude install-skill AlexeyGOblov/visual-attention-audit

# Visual attention audit Two questions, answered with numbers: 1. **How cluttered is this frame, and what draws the eye first?** 2. **Does it hold up among its neighbours, at the size people actually see it?** Both run on plain images. No browser, no network, no model weights. Dependencies are numpy, scipy and Pillow. ## Verify before you trust it ```bash python scripts/clutter.py --selftest ``` Three checks must pass: | Check | What it proves | Expected | |---|---|---| | Frequency tiling | the steerable pyramid is built correctly | mean 0.999994, max deviation 1.3e-05 | | Monotonicity | clutter rises with item count (0, 8, 40, 200 items) | FC 1.2 / 4.4 / 11.0 / 20.1 | | Extremes | a blank frame is not like colour noise | FC 1.2 against 19.7 | ## Mode 1: measure frames ```bash python scripts/clutter.py frame.png [more.png ...] --width 190 --json out.json --maps maps/ ``` `--width` rescales to the width the design is actually displayed at, before measuring. This is not cosmetic. A 1200 px design shown at 190 px in a grid is a different stimulus: text that carried the message at full size stops existing at thumbnail size. Measure at every size the design is really seen at, not only at the size it was authored at. Typical display widths worth measuring, in a shopping or gallery context: | Context | Width | |---|---| | Grid thumbnail, phone | 170-190 px | | Grid thumbnail, desktop | 220-260 px | | Detail view, phone | 390 px | | Detail view, desktop | 500-700 px | Ou