I am working on the AIY vision kit and trying to implement object detection in a video stream.
I have tried following the method posted on the AIY website(https://aiyprojects.withgoogle.com/) but seem to be facing a problem when I try to convert the model using the bonnet model compiler.
This is a brief summary of the steps I followed:
- create annotations for each image in the training set in xml format(256x256).
- convert all the images to 256x256.
- create the tf records.
- downloaded the base model of ssd_mobilnet_v1_coco.config and made the required changes.(set num classes = 2, and path to train and val tf records).
- trained the model
- exported the frozen graph to binaryproto format.
These 5 steps are can be seen by following this link,the same images were used but resized.(https://medium.freecodecamp.org/trackin … c86419225e)
all the functions such as create tf record and other related functions can be found in this git repo https://github.com/bourdakos1/Custom-Object-Detection
After I export the model, I get the following error while using the bonnet model compiler.
Any help on this would be highly appreciated