Counting in a heap of fruits

Hello friends,

I want to build a model which can count the number of coconuts in a heap. I am planning to manually collect data for this. I have found this interesting paper. Few questions I have.

I see that there is no way I can detect coconuts as a heap completely covers many coconuts inside. So I want to consider it as a regression problem.

  1. Which angle of the heap should I collect data? The top view (of the heap) seems to consider the height of the heap as well, but not easy to collect.
  2. Maybe I should start with a smaller fruit like lemon so that I can easily collect data and train and later do transfer learning on coconuts?
  3. A square image of 5 MP should be good enough for collecting? Or is it an overkill?
  4. Any place where I can get pre-trained weights to start with?

Please feel free to express your thoughts.

1 Like

Surely the CoCo dataset :slight_smile:

Seriously though, assuming you have the luxury of time and a bigger model, perhaps an object detection solution would work well?

1 Like

In a heap (pyramid kind of), many coconuts are hidden deep inside. There is no way to detect each coconut.

I’ve really no idea whether this has any relevance, but maybe it’s worth trying;

1 Like