We’ve been doing some testing with the Movidius stick and Tensorflow, running into some similar issues throughout the forums. We’ve stitched individual ideas and pieces from different forum posts + resources into a general workspace repository for those wanting all the tooling in a single space, available here:
https://github.com/Technica-Corporation/TF-Movidius-Finetune
The main things we’ve learned that may be of interest are transfer learning/fine-tuning pre-trained models on your own custom image data. We’ve been focused on the MobileNet model architecture with which we’ve been having difficulty generating similar results. Open up for any suggestions to solve what we’ve been running into.
