[core] Discrepancy in Model Inference Results Between Aidge and ONNX Runtime

What commit version of aidge do you use

aidge_backend_cpu 0.4.0 aidge_core 0.4.0 aidge_onnx 0.3.0

Problem description

Description:

I have observed significant differences in the inference results when using the same ONNX model file with Aidge and ONNX Runtime. The input image and the inference outputs from both frameworks are shown below:

Script to reproduce error heretest_model_segmentation_simple.py.

How can we identify where the model is going wrong?

Can we compare Intermediate Outputs from different layers of the model in both frameworks?

I hope this can help identify at which stage the outputs start to diverge.

I use the same preprocessing and post processing for both frameworks.