[Bug]: Models YOLOv8s and YOLOv9c started to crash on NPU with newest versions of OpenVINO and benchmark_app

[dmichalo]

OpenVINO Version

2024.4.0

Operating System

Other (Please specify in description)

Device used for inference

NPU

Framework

None

Model used

YOLOv8s, YOLOv9c

Issue description

OpenVINO's benchmark_app started to crash on models YOLOv8s, YOLOv9c running on NPU. These models work without any issues on CPU and GPU. There were working as well when was converted to IR format with older versions of OpenVINO.

At the moment I identified these 2 YOLO models, but it it possible that other are affected too. The problem is observed with OpenVINO versions 2024.4. and 2024.3. When version 2024.2 was used, benchmark_app does not throw any errors.

OS: Ubuntu 22.04.4 LTS

Step-by-step reproduction

1. Download and covert YOLO model:
   from ultralytics import YOLO pt_model = YOLO("yolov8s.pt") pt_model.export(format='openvino', half=True)

2. Use model:
   benchmark_app -t 10 -d NPU -inference_only=false -hint throughput -m yolov8s.xml

Relevant log output

Screenshots attached.

Issue submission checklist

• I'm reporting an issue. It's not a question.
• I checked the problem with the documentation, FAQ, open issues, Stack Overflow, etc., and have not found a solution.
• There is reproducer code and related data files such as images, videos, models, etc.