Image classification- Why am I getting vastly different results training on Tensorflow vs Pytorch?

Question

I am training an image classifier for a robot that detects "blocked" vs "free" paths. Using the same data on pytorch gives >0.98 accuracy on validation da