<< return to Vizycam.com

Adding new tensor flow models

Vizy applications
#1

Hi all,

I’m from Australia so wanted to add a new tensor flow model to classify birds we have here

I used this classifier: https://tfhub.dev/google/aiy/vision/classifier/birds_V1/1
Then updated /usr/local/lib/python3.7/dist-packages/kritter/ with my edits from the git repo above

Made a copy of the TensorFlow app and modified it to use my BIRDS classifier instead of COCO

Firstly it didn’t like that the background id wasn’t 0 (so I removed this from the label map)
but now I’m getting an error:

google.protobuf.message.DecodeError: Truncated message.

Process has exited

Does anyone have recommendations on how I can use this model with the camera?

#2

Hello,
Vizy uses a detection network (ssd mobilenetv2) for detecting birds/objects. A classifier network has a different architecture and won’t drop in (so to speak) to the existing detection network. (sorry!) :slight_smile:

We are going to release a new version of the detector hopefully by August. It will be more accurate and easier to program.

We’ve promised to provide a European birds network. I think we should include Australia :slight_smile: At any rate, we hope that training your own through a cloud platform like Google Colab will be something someone with moderate technical abilities can accomplish.

Edward

#3

it looks like it should work? There is also a TF Lite version of it

image

#4

Ahh I see, so the SSD part is what draws the box and that is missing…