ONNX image segmentation

This guide shows how to use some object detection models (exported to ONNX format) in Breeze.

The supported object detection algorithms via ONNX, currently are YOLOv4, YOLOv5, YOLOv8, YOLO11 and Faster R-CNN. To utilize this feature add a segmentation node to the Analyse Tree, Deep learning segmentation

Select your pre-trained ONNX model type in the Model Type drop-down and browse to and select the model file, in this case, a Faster R-CNN model file and segmentation.

A Label classification node is automatically added when adding the machine learning segmentation. Add a new line separated class file to the Segmentation label node. May be in either .txt or .names files.

Click Apply Changes to evaluate the segmentation, go to the Table tab to see the result.

To train a custom object detection model we recommend using PyTorch and one of the available repositories for training specific version of a network, here is an example which trains a YOLOv5 model with custom data: https://colab.research.google.com/github/Prediktera/Train-Onnx-Image-Segmentation/blob/main/YOLOv5/YOLOv5-colab.ipynb