Track operations with Jobs

1. What is a Job?

When using Picsellia to advance your Computer Vision project, you may need to perform tasks that require significant time and computing resources. These tasks, such as importing a large annotation file, training a model, or enhancing your Dataare referred to as Job on Picsellia.

To ensure that users are aware of the progress of these Job, you can use the Jobs tab to track their progress and access related logs. There are three types of Picsellia Job:

• Processings: These are scripts that utilize Picsellia to apply pre-annotation or data augmentation to Dataset. Both public and user-created processings can be accessed in the Jobs tab.
• Trainings: These are model training completed on Picsellia's infrastructure. Both public models and user-imported models and training scripts can be accessed in the Jobs tab.
• Internals: These are Picsellia's native object manipulations, such as annotation imports or Dataset forks.

Trainings and Processings can each run on the Compute Connector of your choice, and their consumption is reflected in your DPU & MPU usage.

2. Job details

For each Job, you'll retrieve several pieces of contextual information (from left to right):

• The status of the Job (Success/Failure/Running)
• The name of the Job (Clicking on it will display the logs of this action)
• The user who triggers this Job
• The Picsellia object affected by this Job (Clicking on it will show the related object e.g.Dataset Version)
• The duration since the Job has been triggered
• The duration of the Job to be completed

As mentioned, by clicking on the Job name, you will access a detailed view of the related Job.

In addition to the Job details already displayed in the Jobs tab, this detailed view will let you access to:

• The Job logs if available. For example, internal Job are not generating logs. Regarding non-public processing & training imported by users, it depends on the logs generated in the dockerized script imported to the Picsellia model or processing private registry.
• The Job ID (for SDK use)
• The possibility of relaunching the Job thanks to the Re-run button in case of a Processing (top right corner)