DEVICE-TEST: multi-photo tip — take 2-3 pattern shots for visual-ID (app fuses views; single shot ~28% top-1, multi-view is the reliable path)

a376a2d59a06f261d3006d80a1f087bb9feac204 · 2026-07-07 13:06:17 -0700 · Steve Abrams

Files touched

M DEVICE-TEST.md

Diff

commit a376a2d59a06f261d3006d80a1f087bb9feac204
Author: Steve Abrams <steve@designerwallcoverings.com>
Date:   Tue Jul 7 13:06:17 2026 -0700

    DEVICE-TEST: multi-photo tip — take 2-3 pattern shots for visual-ID (app fuses views; single shot ~28% top-1, multi-view is the reliable path)
---
 DEVICE-TEST.md | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/DEVICE-TEST.md b/DEVICE-TEST.md
index 620a351..8f63923 100644
--- a/DEVICE-TEST.md
+++ b/DEVICE-TEST.md
@@ -37,6 +37,10 @@ measure paths that only a real device can exercise.
 - [ ] **Visual-ID path:** with the mfr# blank, shoot the **front pattern** → it should match by
       design ("🔍 Visually matched <title>" once the re-embed is done). If it auto-matched,
       the **first Push asks "tap Push again to confirm"** — that's the safety gate; a 2nd Push writes.
+- [ ] **Multi-photo tip (matters):** for visual-ID, take **2-3 pattern shots** from slightly different
+      angles/distances before it resolves — the app **fuses all views** (a pattern matching across
+      several shots outranks single-shot noise). Single handheld shots are only ~1-in-3 top-1;
+      multiple views is the reliable way to ID by design. Same applies to the Add-mode dup warning.
 - [ ] Edit the mfr# after a match → the target should reset ("mfr# changed — re-find") so you never
       push to a stale product.

← de2aca3 robustness_fusion: note CDN rate-limiting makes the A/B unre · back to Dw Photo Capture · Fix aug-training crash (cannot write empty image): guard aug 27b572c →