AI Metal Assembly Image Comparator - 08/07/2026 09:42 EDT

Open

Prize:

$1,500 USD

Entries Received:

13 days, 23 hours remaining

I need a small, accurate program that accepts two medium-resolution JPEG photographs of a finished metal assembly and automatically presents them side-by-side, with every visual difference clearly highlighted on the right-hand image. The purpose is simple: help our fabrication team spot mis-welds, missing brackets, or other subtle manufacturing deviations without having to inspect the parts manually. Here is what matters most to me: 1. The tool must read standard JPEGs shot between 720p and 1080p—you can assume consistent lighting but plenty of reflections and shiny edges typical of stainless steel. 2. Output should be a single composite image (or screen) that shows both originals next to each other; regions that differ are accented visually so a technician can locate them in seconds. Feel free to decide whether to use bounding boxes, semi-transparent color flashes, or other intuitive cues, as long as the comparison view remains uncluttered. 3. Speed is valuable on the shop floor, so an algorithmic approach built with OpenCV, scikit-image, Pillow, or comparable computer-vision libraries is preferred over heavyweight cloud inference. Python is ideal, but I’m open to C++, .NET, or a lightweight desktop executable if performance warrants it. 4. The detector should ignore irrelevant noise such as slight lighting variations but catch dimensional or component changes down to a few millimeters in scale. Deliverables: • Compiled application (or runnable script) with a straightforward interface: select Image A and Image B, press “Compare”, receive the highlighted side-by-side result. • Source code with clear comments so my in-house developer can tweak thresholds later. • A brief read-me describing dependencies, installation, and how to add sample images for testing. Acceptance will be based on a small test set of actual shop photos I’ll provide; all genuine mistakes must be marked, and no more than 5% false positives may appear. Let me know which libraries you plan to leverage and how you’ll tune for reflective metal surfaces, and we can get started right away.