0

I want to detect if there are objects within a boundary.

The boundary is white rectangle, and easily identifiable (for a human).

However, the position of the boundary is not fixed.

The objects are small, and usually only 1 or 2 present in the boundary - but are visible.

The sample images are only labelled with 1 if any object is in the boundary, 0 if not. In particular, I don't have the boundary as a label.

What is a good architecture for classifier of such images? Are layers of CNN + MaxPooling my best bet?

KalEl
  • 8,978
  • 13
  • 47
  • 56
  • It would be great if you could post a few sample images, so that we can understand the task. – Ganesh Tata Jun 19 '19 at 11:14
  • 1
    Unfortunately I couldn't share an image, because it's for a protected application. However, I can share what worked. Resnet50 was taking too long to converge, plus it needed hacks to use just the great and depth channels. The architecture I resorted to instead was shallower, with three towers, Broad Features + Intensity + Depth. Intensity and Depth had id and convolutional residual blocks. Also used the output intermediately to make the training even faster. Took 400 epochs to train from scratch to yield 98+% accuracy on a (balanced) validation sample - which is pretty good. – KalEl Jul 13 '19 at 15:00
  • That's awesome! Since you can also answer your own question, it would be great if you could write your comment as an answer, so that others are aware of what really worked for the given problem statement. Thank you! – Ganesh Tata Jul 14 '19 at 07:40

0 Answers0