Developer Builds Custom Julia Image Processing Library After Studying ML Data Augmentation

A software developer created a personal image processing library in Julia while studying machine learning, motivated by a desire to independently implement data augmentation techniques. Starting with image rotation, which he mentally worked out during a 2.5-hour shift at his manufacturing job, he initially produced a slow solution taking around 1,000 ms per operation. After discovering the inverse mapping and bilinear interpolation technique, he optimized his rotation function to outperform the existing Augmentor.jl library. He subsequently implemented seven image transformation operations — including distortion, flipping, scaling, shearing, and zooming — on both CPU and GPU. The project, intended for machine learning and general image processing use, was built largely from first principles with only selective reference to established methods.