Dear PyTorch Community Members and Experts,
We have been doing some research that involved using Deep Learning Networks to classify images of Chest X-Ray for various diseases (e.g. COVID-19, Viral Pneumonia, SARS, MERS, etc.). To our surprise, we noticed a big gap between the results from MATLAB and PyTorch. We used all the parameters the same during Training or Testing (e.g. Train and Test Data, Learning Rate, Number of Folds, Optimizer, Cost function, etc.). For the same models (e.g. VGG_16), MATLAB was performing significantly better, more than 5% in some cases, which is huge in the extreme (e.g. 90% is much better than 85%). Does anyone have any idea what could be the issue here? We prefer PyTorch results for our paper but MATLAB is performing better.