[Bug] DirectML performs the same as CPU and using CPU > GPU

[aihkla]

Prerequisite

• [X] I have searched Issues and Discussions but cannot get the expected help.
• [X] The bug has not been fixed in the latest version(https://github.com/open-mmlab/mmpose).

Environment

Windows using ONNX runtime

Reproduces the problem - code sample

import cv2 import time from rtmlib import Wholebody, Body, RTMPose, RTMO, draw_skeleton import onnxruntime as ort

device = 'dml' # cpu, cuda, mps, dml backend = 'onnxruntime' # opencv, onnxruntime, openvino

openpose_skeleton = False

ort.set_default_logger_severity(0)

Reproduces the problem - command or script

Device is set to DML, but GPU occupy only 2-3% and CPU usage is over 60%

Reproduces the problem - error message

Device is set to DML, but GPU occupy only 2-3% and CPU usage is over 60%

Additional information

DML supposed to run faster than CPU only mode and occupy more GPU