During March 2026, contributed to the opencv/opencv repository by developing a feature that preserves the axis attribute during Gather-Cast fusion within the DNN module. This work focused on enhancing ONNX model compatibility and stability by ensuring axis metadata is retained throughout the fusion process, aligning OpenCV’s DNN behavior with ONNX semantics. The implementation required a strong understanding of C++ and CMake, as well as expertise in computer vision and deep learning frameworks. By addressing edge-case failures in model deployment, this contribution improved the reliability of ONNX-backed inference workflows and demonstrated proficiency in DNN graph fusion and metadata management.