Choosing labels based on their names for classification models

[jcchr]

Summary

Model API - used by geti-sdk to perform inference, returns results based on a list of labels which is sorted alphabetically - whereas geti-sdk works on a list ordered like it is in a model itself. In some cases order of labels might be different in both libraries - which leads to problems like the one described here : open-edge-platform/geti#1578

Correction is to establish labels returned by model based on their names instead of using their location on a list. Change should be done for classification models only - as a change related to sorting labels was introduced only for this kind of models.

How to test

Execute inference on a model - checking first if labels in a model (check the .xml file with models - list of labels is located at the end of it) are not sorted alphabetically. Check if results of inference are ok.

Checklist

• I have tested my changes manually.​
• I have added tests to cover my changes.​

License

• I submit my code changes under the same Apache License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below).

# Copyright (C) 2025 Intel Corporation
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions
# and limitations under the License.

[Copilot]

Pull request overview

This PR fixes a label ordering mismatch between geti-sdk and ModelAPI. ModelAPI returns inference results with labels sorted alphabetically, while geti-sdk uses the model's original label order. This inconsistency caused incorrect inference results when the orders differed.

Key changes:

• Modified label processing in ResultsToPredictionConverter to sort labels alphabetically before mapping inference results

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[leoll2]

Model API - used by geti-sdk to perform inference, returns results based on a list of labels which is sorted alphabetically

If confirmed, this is a rather strange behavior. Can you provide a reference to the ModelAPI code that sorts the labels alphabetically?

[leoll2]

The purpose of this test case was to check an edge case where the label names are very similar and conflict after escaping (both foo bar and foo_bar become foo_bar). The new label names foo bar1 and foo_bar2 do not trigger the intended edge case.

[leoll2]

What's wrong with get_label_by_idx?

[maxxgx]

I remember that MAPI does not handle hierarchical classification in a consistent way, so we could not rely on the label indices. See related issue: open-edge-platform/geti#402.

Therefore, it is correct to use the label name for classification tasks. However, this means that we might get ambigous results if label are poorly named. For example, foo bar and foo_bar, spaces are not supported by either MAPI or OV AFAIK, resulting in a name collision.

[maxxgx]

I remember that MAPI does not handle hierarchical classification in a consistent way, so we could not rely on the label indices. See related issue: open-edge-platform/geti#402.

Therefore, it is correct to use the label name for classification tasks. However, this means that we might get ambigous results if label are poorly named. For example, foo bar and foo_bar, spaces are not supported by either MAPI or OV AFAIK, resulting in a name collision.