OpenCV and Dlib settings

I was super excited to try out the image_processing components with my 4 Hikvision cameras, but am not too happy with the performance of both these platforms. I get lot of false positives, reducing their value for any automation. I am guessing that it is largely due to my sub-optimal settings and things can be improved with the right settings. I am sure many of us are in the same boat and can use some guidance. Here are my configurations (more can be found on my repo):

image_processing:
  - platform: opencv
    name: OpenCV
    source:
      - entity_id: camera.porch
      - entity_id: camera.patio
      - entity_id: camera.backyard
      - entity_id: camera.driveway
    classifier:
      body: /home/homeassistant/.homeassistant/classifier/haarcascade_fullbody.xml
      neighbors: 5
  - platform: dlib_face_detect
    source:
      - entity_id: camera.porch
      - entity_id: camera.patio

I have tried them with both high-res feed (2688x1520) that is resource intensive and low-res (640x360) without much difference. If you have these components working well for you, kindly share your configurations and the camera stream profile (resolution).