Worked on the opencv/opencv repository to deliver a targeted performance optimization for masked accumulate operations on 4-channel images. Developed new SIMD paths for the cv::accumulate() function using C++ and OpenCV Universal Intrinsics, specifically supporting both CV_32FC4 and CV_8UC4 to CV_32FC4 conversions. The implementation included robust regression tests and dedicated performance benchmarks, quantifying speedups up to 3.5x for common 8UC4 workloads. Maintained correctness by preserving scalar fallback for non-multiple input lengths and enhanced the test infrastructure to ensure stability. The work demonstrated strong skills in SIMD programming, image processing, and performance optimization within a complex codebase.