Hi Mohan, thanks for the explanation. Suddenly i see your point of view! Trying to redo P3 right now though i have already cleared Term 1. Felt that there was still something i am missing out in P3 while i was building and training the model.
Just another question and feel free to explain only when you are free.
From your post i understand that you had used L/R images and flip images. That makes 6 images and choosing them randomly. Based on this method of feeding data to your model, wouldn’t it be more bias to feed L/R images (66%) than center image? If so, doesn’t it affect the training outcome.
I’ve tried your method of feeding through python generator and also feeding L/R images 25% of the time with center image 75% of the time where it seemed to drive better. Any opinions on that?