During training several snapshots of model are saved and could be retrieved after the training process is finished. This can be useful to understand training dynamics and select the best available model for further usage.
Suppose that we have run training process and corresponding task was created and finished.
In a context menu assotiated with the task there is a "Checkpoints" item that give you access to all the checkpoints that were created during the training process
After clicking on the "Checkpoints" item the checkpoint list will be available. Select the checkpoint you want to use, and specify the corresponding title. After that click "Create" button
In "My Models" tab the model with specified name is available now.
You can treat this model exactly the same way as other available models
After the training process is finished, a model which corresponds to the latest checkpoint is automatically added in "My models" tab
For example, in a case of UNet training, checkpoints #3, #7, #10 are added to "My models" and the segmentation results are visualised
Here is segmentation results of a model from checkpoints #3
Here is segmentation results of a model from checkpoints #7
Here is segmentation results of a model from checkpoints #10
You can see that as training progresses the results are getting better
Unused checkpoint cleanup¶
In a lot of cases, models might occupy a significant amount of disk space, it's important to remove the checkpoints that will not be useful in a future.
Checkpoints are associated with a particular training task. In order to remove unused checkpoints, you need to select "Clean unused checkpoints" in a context menu of specified task
Before checkpoints are removed you will see the following warning