Does a 100% white-edge score prove the whole product background is pure white?

What our actual test measured

On October 9–10, 2026, we generated a 1254 × 1254 headphone packshot through a local CartVisual candidate using a project-owned demonstration reference. The original had a near-white, paper-like background. The website’s twelve-pixel edge sample reported 10.7% pure white. After its near-white correction, 657,642 pixels were changed and the sampled edge reported 100.0% pure white.

The downloaded corrected PNG still showed near-white texture in the closed area inside the headband. We therefore did not mark the entire background as strictly pure white. This is one actual local test, not a production-wide success rate, a customer result or evidence that a marketplace accepted the image.

The denominator in this archived example is 59,616 sampled pixels: the 1254 × 1254 canvas minus its 1230 × 1230 interior. The original has 6,385 exactly opaque-white samples, or about 10.7%; the corrected file has 59,616, or 100.0%. The 657,642 changed pixels count the full-image difference, not the edge-score denominator. These describe the retained files, not a new generation or a new marketplace test.

The current perimeter check derives its band from one percent of the shorter image side, rounded down with a minimum of one pixel; this image therefore uses 12 pixels. Do not describe every future upload as a fixed twelve-pixel test. Neither the band count nor the whole-image change count identifies which changed pixels were truly background.

Compare the actual files

Name the region behind each metric

Do not turn every near-white pixel into white to manufacture a passing number. That can erase white product surfaces, reflections or useful edge information. A product/background mask needs its own accuracy review.

Use a scoped approval statement

A truthful record for this test says: outer sampled band passed after correction; closed headband background still needs review; product identity and destination rules checked separately. For your own image, identify all holes and enclosed spaces before approving a strict background. A higher percentage alone does not tell you whether the correction damaged the merchandise.

CartVisual’s tested correction was not a complete semantic cutout. The general lesson is to describe what a metric measured and inspect what it omitted. A marketplace or customer may have different background requirements; follow the actual destination rather than treating this case as a universal rule.

Current CartVisual correction controls include an outer near-white correction and selection of an enclosed near-white region by point or source-pixel coordinates, with comparison and undo. This historical outer-only result does not mean the current interface lacks that additional step. The operation still follows a selected near-white region; it is not semantic product/background segmentation and does not certify a safe selection on white merchandise.

For a new file, identify each genuinely background opening before making a local correction, compare the result with the preserved original, and reject changes to merchandise. If the boundary is uncertain, continue in an appropriate reviewed cutout workflow. Use white-product boundary review to protect actual shading. A region-specific record is more useful than raising a number while leaving an uninspected hole.