Train a caffe model using DIGITS and deploy it in DeepStream SDK

Hello, i want to train a face detector caffe model, so i can use it in Deepstream. I’ve gathered a dataset based on open images dataset v4. Im trying to use the resnet.prototxt located in deepstream_sdk_v4.0.1_jetson/samples/models/Primary_Detector_Nano and the associated weights. My question is, what are the steps i shoul follow? Shoul i resize all of my images in my dataset in DIGITS? once i have the caffe model, can i use it directly in deepstream? do i have to generate a .engine model?


The model can be re-trained with DIGITS directly.