Gravio supports TensorFlow and TensorFlow Lite models. TensorFlow Lite are lighter and therefore strain the computer less, however they are also less accurate. It highly depends on your situation and the quality of the image if the TensorFlow Lite model is enough. Generally, if your equipment is powerful enough we recommend to use the TensorFlow models. For the Gravio Hub, we recommend TensorFlow Lite.

In this section, we learn how to create a TensorFlow 1.x model file as a TensorFlow model file that can be used with Gravio.

Refer to this page to learn how to create a TensorFlow Lite model file.

The Models Overview

In this view, you can see all the models that are available for deployment. Please note, if your list is empty, you can fill it either by uploading your own models, which we will show further below or, you can use a few pre-trained models by logging in under “License” using your Gravio username and password that you used to sign up. If you log in with your credentials and open this screen, you will see a list similar to the below. Note that the available default models depend on your license package.


 
Let’s look at the various ways of adding your models:

Download a pre-trained model file from Gravio.com

Click the Download Model File icon to download the model file from gravio.com to your Coordinator. You will need to download the model file in order to edit it.

After downloading the model file, additional buttons will appear on the right:

Export: downloads the model file to your local machine.
Copy: duplicates the model file with the configuration information changed.
Edit: edits the setting information of the model file.
Delete: to delete the model file from your Coordinator.

Exporting a model file

You can click the export button to download the model to your local computer. Models that have been downloaded in this manner can then be re-uploaded again using the “upload” button on the top right.

Duplicate model a file

You can duplicate a model file in order to change basic parameters.

Click the Duplicate Model File button to display the Duplicate panel.
Now you can change the model name, method, output format, detection value, and confidence threshold.
The model name is automatically appended with “Copy_”.
Click the “Save” button to duplicate the model file.

Editing a model file

To edit the parameters of a model file, click the Edit Model File icon to display the edit panel.
Now you can change the model name, method, output format, detection value, and confidence threshold.
Make sure you have a new name for your newly created model file.
Click the “Save” button to duplicate the model file.

Creating a new model file from scratch

To upload a TensorFlow model file that you have created from scratch, click on the “Create Model” button on the top right and pick “TensorFlow Model” from the dropdown. To learn how to create the packages, please follow the instructions on TensorFlow.org.

Once selected, the following popup will appear:

Files to prepare

To create a TensorFlow model file, prepare the following files:

File
TensorFlow metafile
TensorFlow pb file

Items to be registered

Item name Description
model name set in alphanumeric characters.
TensorFlow metafile Select the file you prepared.
TensorFlow pb file Select the file you prepared.
Method Select “Count” to output numbers or “GroupBy” to return a string.
Form of output Select JSON to output detailed information, or select numeric to output only values.
Include detection values Select to include detection values.
Confidence threshold Set the confidence threshold in the range between 0.0 and 1.0.

Then upload them in the respective fields.

Was this helpful?

Yes No
You indicated this topic was not helpful to you ...
Could you please leave a comment telling us why? Thank you!
Thanks for your feedback.