Why the results of 'performance' mode is worse than that of 'balanced'

[awesome-wflow]

Hi,

I tested the "demo.jpg" with 'performance' mode and 'balanced' mode, and found that the result from 'performance' mode is worse? Is it correct?

I attach some inference codes I used bellow and inference results from different modes.
Could you please tell me whether there is any bugs in the codes?

Inference codes:

import cv2

from rtmlib import Wholebody, draw_skeleton

device = 'cuda'  # cpu, cuda, mps
backend = 'onnxruntime'  # opencv, onnxruntime, openvino
img = cv2.imread('./demo.jpg')

openpose_skeleton = False  # True for openpose-style, False for mmpose-style

wholebody = Wholebody(to_openpose=openpose_skeleton,
                      mode='balanced',  # 'performance', 'lightweight', 'balanced'. Default: 'balanced'
                      backend=backend, device=device)

keypoints, scores = wholebody(img)

# visualize

# img_show = np.zeros(img_show.shape, dtype=np.uint8)
img_show = draw_skeleton(img, keypoints, scores, kpt_thr=0.5)


cv2.imwrite("./demo_result.jpg", img_show)

setting in wholebody.py:

class Wholebody:

    MODE = {
        'performance': {
            'det':
            '/mnt/new_nas/zhenyuxie/MCA-project/datas/rtmlib/ckpts/yolox_m_8xb8-300e_humanart-c2c7a14a/end2end.onnx',  # noqa
            'det_input_size': (640, 640),
            'pose':
            '/mnt/new_nas/zhenyuxie/MCA-project/datas/rtmlib/ckpts/rtmw-dw-x-l_simcc-cocktail14_270e-384x288_20231122/end2end.onnx',  # noqa
            'pose_input_size': (288, 384),
        },
        'lightweight': {
            'det':
            'https://download.openmmlab.com/mmpose/v1/projects/rtmposev1/onnx_sdk/yolox_tiny_8xb8-300e_humanart-6f3252f9.zip',  # noqa
            'det_input_size': (416, 416),
            'pose':
            'https://download.openmmlab.com/mmpose/v1/projects/rtmw/onnx_sdk/rtmw-dw-l-m_simcc-cocktail14_270e-256x192_20231122.zip',  # noqa
            'pose_input_size': (192, 256),
        },
        'balanced': {
            'det':
            '/mnt/new_nas/zhenyuxie/MCA-project/datas/rtmlib/ckpts/yolox_m_8xb8-300e_humanart-c2c7a14a/end2end.onnx',  # noqa
            'det_input_size': (640, 640),
            'pose':
            '/mnt/new_nas/zhenyuxie/MCA-project/datas/rtmlib/ckpts/rtmw-dw-x-l_simcc-cocktail14_270e-256x192_20231122/end2end.onnx',  # noqa
            'pose_input_size': (192, 256),
        }
    }

Here is the result from 'performance' mode

Here is the result from 'balanced' mode

[Caiuszhao]

Maybe the meaning of 'performance' refers to 'speed' rather than 'quality'?