Performacne Comparison of Object Detection implementations in MATLAB and PyTorch

Objective

The objective was to profile deep neural network implementations for instance segmentation architectures both in MATLAB (Mask RCNN) and PyTorch (detectron2).

Methodology

• Developed scripts for training models in MATLAB & PyTorch with same architecture & training options and log time for various somputation steps such data loading, forward pass, backward pass, roi generation etc.
• Compared the profiling, identiied bottelnecks and presented findings to development team for performance improvement.

Tools & Technologies

Computer Vision, Instance Segmentation, Mask R-CNN, detectron2, PyTorch, MATLAB, Linux, GPU