Google Object detection model lifespan for Edge Devices

Question

I recently got to know about a strange thing: The Object detection model has a Lifespan of 18 months on both Edge devices / Offline model exported as Tflite for Android and On cloud-hosted model. I want to know the reason behind this.

Thank you in advance for your help and support.

Following are the links I came across and found this issue:-

1. Models for Edge Devices:- https://cloud.google.com/vision/automl/docs/train-edge
2. Models on the cloud:- https://cloud.google.com/vision/automl/object-detection/docs/models

Answer 1

The 18 months lifespan defined is for old and stale models, that is if the model has not been used within that time it will be deleted as it will no longer be working as an AutoML model rather it will serve as a simple dataset.

The reason for the defined lifespan is, if the model is inactive it may not operate properly, that is it may not provide accurate results. Beyond 18 months Google will serve the model at best effort basis.

To keep the model in working state after the defined lifespan, we need to retrain the model by providing both new and old data.

AutoML creates a new model each time you start training. It creates a new model from the beginning so all the information needs to be used.