Keypoints Detection

I am trying to detect keypoints in grayscale images. The problem that I have is that my keypoints are not consistent for each image. So image1 has 10 keypoints and image 2 has 20 keypoints and so on.
In total I have 32 keypoints max but as I said each image has varied number of keypoints.

So how can I modify my model to accomodate varied number of keypoints?

self.fc3 = nn.Linear(400, 32*2)

Please guide me

@ptrblck can you please help