Encoding works fine. Decoding fails ~50% due to camera color distortion, lighting variance, and compression artifacts.
Tech: Kotlin/C++, OpenCV, Reed-Solomon over GF(2^8), LZMA compression.
Main question: is this fundamentally broken with consumer cameras, or solvable with better calibration/ML?
Apache 2.0 licensed. The Reed-Solomon implementation might be useful even if the barcode concept doesn't pan out.
Even an examples directory in the repo with some samples would have helped.