Training a detectron2 model on intel neural compute stick2

Question

Any one knows how to train faster_rcnn_R_50_FPN_3x model which is pytorch-based model from detectron2 models zoo on intel neural compute stick2 device ?
I've already installed openvino-toolkit and run correctly one of their demos on my device.
I've trained the faster_rcnn_R_50_FPN_3x on a custom data set with detectron2 on google-colab-gpu and now I have to train it on the intel neural compute stick2 vpu to compare the AP results

Answer 1

You cannot train models on the Intel Neural Compute Stick 2 (NCS2). NCS2 is to be used with OpenVINO™ toolkit. OpenVINO™ toolkit consists of two main components, which are Model Optimizer and Inference Engine. The Inference Engine MYRIAD plugin has been developed for inference of neural networks on Intel® Neural Compute Stick 2. Have a look on this OpenVINO™ Toolkit Workflow.