Detectron2 on Intel integrated GPU Iris Xe

[morussk]

Hey,
Does anyone know why wouldn't Detectron2 run on integrated Xe graphics card using OpenVINO?

It does work fine with Intel CPU but the moment I change to GPU it crashes without error on 

compiled = core.compile_model(model, "GPU.0")

When I try to load YOLO11 model on Intel GPU it works totally fine as well so I am suspecting there is something to do specifically with Detectron2 model/architecture but havent been able to understand it.

Here is what my AI assistant came up with, perhaps you can confirm/deny/fix:

Confirmed Bug

This is a bug in OpenVINO's Intel GPU plugin that causes a segmentation fault when compiling Faster R-CNN models. The Intel GPU OpenCL kernel compiler crashes when processing the model's dynamic operations (ROI Align, NMS, Gather, etc.).

Summary Table

Model Intel GPU Crash Type Evidence

Simple ReLU	Works	-	-
YOLO11n	Works (7ms, 144 FPS)	-	-
Faster R-CNN	Crashes	SEGFAULT (0xC0000005)	Access violation in compile_model

Optional: Report to Intel

You could report this bug to Intel with this information:

GitHub Issue: https://github.com/openvinotoolkit/openvino/issues

Report contents:

• OpenVINO version: 2025.4.1
• Intel GPU: Iris Xe Graphics (iGPU), driver 32.0.101.7026
• Model: Detectron2 Faster R-CNN (1966 operations)
• Error: SEGFAULT (0xC0000005) during core.compile_model(model, "GPU.0")
• Note: YOLO models work fine on same GPU

[Wan_Intel]

Hi morussk,

Thank you for reaching out to us.

Could you please share the Detectron2 Faster R-CNN model (.xml and .bin) with us to further investigate the issue?

Regards,

Wan

[morussk]

Hey,
Here is the best way I think I can share .bin and .xml files - my google drive link: https://drive.google.com/drive/folders/1IQVAF4fD302DWDfD_aa4I_bWimFwrlXN?usp=sharing 

Thanks for looking into it.

[Wan_Intel]

Hi morussk,

Thank you for sharing the model with us.

We will further investigate the issue and provide an update here shortly.

Regards,

Wan

[Wan_Intel]

Hi morussk,

Thank you for your patience.

I've downloaded your model and run your model on Intel® Iris® Xe Graphics and Intel® Arc™ 140V GPU. I also encountered the same issue. No issues were encountered while using CPU plugin.

I'll escalate this case to relevant team to further investigate the issue. We will provide an update shortly.

Regards,

Wan

[Wan_Intel]

Hi morussk,

Thank you for your patience.

To further investigate the issue, could you please share the following information with us?

1. Model in original framework
2. Script or code you used to convert the model into OpenVINO IR format

Regards,

Wan

[Wan_Intel]

Hi morussk,

Thank you for your question.

If you need additional information from Intel, please submit a new thread as this thread will no longer be monitored.

Regards,

Wan