Image positioning

Hello, may I ask which algorithm is better for the whole slide? Each label of the glass slide is different, and there will be organizations behind it, which are different. I want to realize the problem of whether the glass slide is placed correctly. Is there any good method?

These are true
image
image

They have different labels and different tissues behind them


These are the wrong ones

Hi @caifan,

It will be easiest if you can guarantee the slide holders are in the same position in the image every time. For example, the second image you posted, the slide holders are rotated by about 45 deg, which will make detecting slide placement much more difficult.

Assuming you can control the position and lighting of the slide holders, there are a couple of methods you might want to try to see what works best:

  1. Basic image classification. The position of the label/slide should be enough to create differences in the representation of the CNN so that you can classify the images differently.
  2. FOMO. This is potentially a good use case for FOMO (or other object detection). If you can determine the location of the label in the image, you can see if it’s within acceptable limits.

@shawn_edgeimpulse Hello, first of all slides position, he is divided into the left and right, is on a plate, camera module is fixed on the arm, let the camera to the left on the right side of the check, and then there is the uneven lighting I looking for ways to solve problem, it will affect the result of classification, FOMO I don’t have to try, this scenario requires accurate classification, 100, I will try the FOMO you mentioned. Thank you for your advice.

@shawn_edgeimpulse Hello, can I use GPU for model training?

Hi @caifan,

GPUs are available for enterprise customers. Please see this discussion: Training speed - audio recognition - NN classifier