System Information

OpenCV python version: 4.7.0
Operating System / Platform: Windows 11
Python version: 3.9

Detailed description

When I run the model attached below with CUDA in OpenCV version 4.7.0 and 4.6.0, the inference time droped from 50ms to 200ms. Related PR is opencv/opencv_zoo#128 (comment)

Model file: palm_detection_mediapipe_2022may.onnx

Steps to reproduce

Run benchmark from opencv_zoo.

cd opencv_zoo/benchmark
vim ./config/palm_detection_mediapipe.yaml
# Change "backend" from "default" to "cuda"
# Change "target" from "cpu" to "cuda"
PYTHONPATH=.. python --cfg ./config/palm_detection_mediapipe.yaml --fp32

Issue submission checklist

• I report the issue, it's not a question
• I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
• I updated to the latest OpenCV version and the issue is still there
• There is reproducer code and related data files (videos, images, onnx, etc)