0

Does feeding in a 640x480 image into a 300x300 SSDMobileNetV2 model reduce accuracy?

I am starting with a 2560x1440 image, which I then split into 12 distinct images of 640x480. The speed of this setup is great. Only 12 images to run through for each frame, but I am wondering about the accuracy.

I want accurate detections more than want speed.

Would I be better off configuring the camera to something square like 2160x2160 and then cropping down to a bunch of 300x300 images?

Would this improve accuracy overall?

Thanks!

Isaac Padberg
  • 200
  • 8
  • 1
    also try the stack exchanges for AI/ML/data science https://stackexchange.com/sites – Christoph Rackwitz Jan 25 '22 at 19:46
  • 1
    Also depends on the content of the image (e.g. object sizes) and how the dnn was trained. Basically best for you would be to use some of your data and evaluate multiple approaches for that domain. – Micka Jan 26 '22 at 00:22

0 Answers0