Missing documentation for "recognition" model type, already implemented in QGIS Deepness

[rojberr]

Hi,

we currently working on integrating a custom recognition model into the QGIS Deepness plugin and we noticed that this model type is already implemented — I can see and use it in the latest version of the plugin within QGIS.

However, the official documentation (Model classes and requirements) does not mention recognition models at all.

The documentation currently lists the following supported model classes:

• Segmentation Model
• Detection Model
• Regression Model
• SuperResolution Model

There is no description or guidelines for recognition models, despite them being present in the plugin already.

As a model creator, I would like to ask:

• What requirements should an ONNX file for a recognition model fulfill in order to work correctly with QGIS Deepness?

Specifically:

• What inputs/outputs/ONNX parameters are expected?

Having this information in the documentation would help developers like me correctly prepare and integrate custom models.

Thanks for your work on this great plugin! I’d love to contribute more effectively once I understand the exact model expectations.

[kapiblue]

In more details, our model performs tree species classification on patches of aerial images (resolution 224x224 pixels).
Maybe another way of framing this question of model integration is whether there is a way to integrate our model into Deepness to perform land segmentation in this patch-based approach.

Looking forward to hearing from you and contributing to Deepness!

[rojberr]

@przemyslaw-aszkowski @bartoszptak @dpieczynski @momaabna @Kleebaue @Saareem @karolmajek @knwin @ziyadsheeba @keepcool13 ;) Let us know, we are a team of >8 people waiting for your response 🚀