topdown_demo_with_mmdet.py runs very slow while inferencer_demo.py runs smoothly

[me7obeast]

mmcv 2.0.1
mmdet 3.1.0
mmengine 0.10.4
mmpose 1.3.2
pytorch 2.0.1+cu118

RTX3060

---

So I tried running a video input with inferencer_demo.py with wholebody, everything ran smoothly as if it's just a video playback. Here is my command.
python demo/inferencer_demo.py path_to_video \ --pose2d wholebody --show

However, when I tried inferencing the same video with topdown_demo_with_mmdet.py with wholebody, it's visually really laggy. I didn't count the fps, but it's clearly unacceptable. Command here.
python demo/topdown_demo_with_mmdet.py \ demo/mmdetection_cfg/rtmdet_m_640-8xb32_coco-person.py \ https://download.openmmlab.com/mmpose/v1/projects/rtmpose/rtmdet_m_8xb32-100e_coco-obj365-person-235e8209.pth \ configs/wholebody_2d_keypoint/topdown_heatmap/coco-wholebody/td-hm_hrnet-w48_dark-8xb32-210e_coco-wholebody-384x288.py \ https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w48_coco_wholebody_384x288_dark-f5726563_20200918.pth \ --input path_to_video --show

All the commands I used are from the official documentary of mmpose ( see the whole-body section ), including the models and the weights used. I found it odd that the second one ran so laggy on a RTX3060. Not to mention that the first one was pretty smooth and both were inferencing the wholebody keypoints. Am I missing something here?