I wanted to start developing a project with image recognition. I want to know the difference between the Intel Movidius Neural Compute Stick and TensorFlow to develop this project Any help would greatly be appreciated. What is the difference between them and what is better for accuracy?
Tensorflow → Model creation, learning, reasoning
Neural Compute Stick → Reasoning Only
【Neural Compute Stick】
Inference is made by converting the learning model generated by Caffe or Tensorflow into what is called "graph file" exclusively for Neural Compute Stick.
An environment with Caffe or Tensorflow installed is mandatory.
You can speed up inference speed even on computers without GPU.
It can be executed if Ubuntu 16.04 / Ubuntu Mate / Raspbian with USB 2.0 port is installed.
It is not suitable for running large weight files.
By inferring on a computer equipped with GPU, it can be made faster.
There is a lightweight version of Tensorflow Lite.
It can be executed Windows / Linux.
"Accuracy" is the difference between the size of the model and the layer that can be handled and the amount of learning.
However, Neural Compute Stick does not support many layers.
Custom layers are not supported.
If you want accuracy, I think you should use Tensorflow.