I have recently trained a face detection model using the very popular implementation of keras yolo model. (https://github.com/qqwweee/keras-yolo3). I have managed to successfully convert this model to a tensorRT optimized frozen graph by loading the model and then taking the graph from Keras’s global graph. I used precision as FP16 and max_segments as 50 while converting the graph.
While loading this graph it takes up a vast majority of my RAM (Around 50%) and as soon as the model starts running the Nano crashes. Could anyone help me understand why this could be happening? I’m powering the Nano using 2A micro usb cable.