search

FarrandTom / social-distancing

:star: An easy end-to-end social distancing implementation created using OpenCV! :camera:
16 stars 4 forks source link

Implementation with other algorithms #3

Open bgonzalezfractal opened 4 years ago

bgonzalezfractal commented 4 years ago

Hi FarrandTom! First of all, thx for uploading this code as opensource contributing in research and development all across the world! I've a similar system that works with Yolo.v3 architecture so I actually detect bounding boxes around the person identified instead of the head and use an approximation to calculate the scale factor for vertical and horizontal dimension once warped. Although I really like your approach since you are detecting faces and can expand this to mask detection, so I have a couple questions:

  1. Have you tried this for masks? I feel it can be an amazing approach but I feel like depending on the internal and external parameters of the camera it could not work on some scenarios
  2. I 've developed a face detection algorithm as well, you start with raw_detections and process them along the way, so if I have the predictions of faces from an input video, how well could I adapt your approach to this different architecture?
FarrandTom commented 4 years ago

Hey @bgonzalezfractal,

Thanks for the comment! To answer your questions:

  1. I have tried this for masks- in fact the real time version we have running with IBM Visual Insights is actually an object detection model trained to perform mask detection. This can then serve the purpose of detecting people's heads so it's a nice double whammy!
  2. I'm not sure I understand your question. Are you talking about processing the detections in real time as opposed to batch processing a video?

Cheers, Tom