Task chaining is no longer supported as a dedicated workflow in Geti v3.0

[yvolokitin]

We want to inform the community about a change to supported workflows in the current release.

Geti™ v2 supported special 'chain' task types that composed a detection stage with a classification/segmentation one. This feature was introduced long ago, when the best available models could only excel at a single task. For example, you could have a detector to localize objects within an image, followed by a specialized classifier to recognize the instance type.

Since then, the state of the art of deep learning models has advanced tremendously; today, detection/segmentation models perform exceptionally well end-to-end, in both accuracy and inference latency. Task chaining is essentially obsolete, so it was removed from Geti™ v3.

Migrating a chained project to v3
If you have a "detection → classification" project in Geti™ v2, we recommend migrating it to a single-task "detection" project in v3. Similarly, a "detection → segmentation" project should be migrated to an "instance segmentation" v3 project. In both cases, follow these steps:

1. Export the dataset from your Geti™ v2 instance in COCO format.
2. In Geti™ v3, choose "Create new project from dataset".
3. Upload the zip archive.
4. Select "Object detection" or "Instance segmentation" as task type, depending on the 2nd (last) task of the original project.
5. In the label selector, select only the labels relative to the 2nd (last) task of the original project. Do not include labels like "No object" and "Background".
6. Click Create and wait for the import to finish.
7. Verify that all media and their annotations are correctly imported.

Feedback
If task chaining is important to your workflow, please share your use case in the comments so we can factor it into future planning.

📖 For the full step-by-step guide, see the v2 → v3 migration documentation