I could really use some help!
I want to use caffe / digits or TF to build a simple trainable single object detector and deploy on the NCS. I am using the AWS preinstalled TF system to train but at this point, I'll use anything or any method that works. The most recent failure being related to "No op in nonMaxSuppresionV3 in defined operations" but there have been multiple errors all through the process over the last weeks. Does anyone have a step by step process from image / object training all the way through to deployment on the NCS that they would be willing to share? Tome and Neal have offered guides but nothing that has fully covers the process especially if you are a newbie.
I have deployed pre-trained nets successfully on the stick ( googlenet , yolo)and I have also successfully trained from images. The piece that always fails is between training and producing the graph using the Movidius tool.
Any help is much appreciated