How to Train an AI Model on AmniSphere Web

Prefer a video instead? Navigate to the video walkthrough at the bottom of this page.

Navigate to the Training Sets tab

After logging in using your Amniscient credentials, navigate to the Training Sets tab. In this tab you can view, edit, search, and delete your existing training sets.

Meet the Training Requirements

Note that in order to train a model your training set:

Must contain at least one object with at least one attribute
Must contain at least one background
All objects within the training set must be collected

Once these requirements are met, the set will appear with the status “Ready for Training”. Click on the training set you’d like to use and click on the “Train Model” button.

Train Model

A sidebar will appear with the available speeds to train your model. Models that are trained at a faster rate cost more credits, but note that the speed of your model training will not impact the accuracy of your models. Select the training speed that best suits your needs and click on the “Train” button. If successful, you will be redirected to the Jobs tab to monitor the progress of your model training.

Monitor Annotation & Labeling Progress

There are two processes that happen behind the scenes before a model is created: first the training set is annotated and labelled, and then the model is trained to recognize your objects. Under the Jobs tab, the training and annotation job you have initiated will show as “In Progress” while it is being processed. If this job is successfully completed, the status will show as “Annotated and Labelled” and the second process will begin. If this job is unsuccessful, the status will show as “Failed” and the second process will not begin.

Monitor Training Progress

Once the model training process has begun, you can track its progress under the AI Models tab. While the model is training its status will show as “In Progress” and once it has completed its status will show as “Trained”. Once the model is trained, you can click on the model to view the objects within the model and the associated model ID. These objects will now be detectable through the mobile app if this model ID is chosen.