Object detection in industy

Recently I have a work of detecting the box and determining its size on a moving conveyor. So can I use the technique like object detection, if I can, how many pics do I need to train the network. By the way, Collecting a large number of pictures in a real scene is cumbersome. is there a way to collect a few pictures to get good results? Thanks.

1 Like

@Roger_Jin, The good thing about working with images is that quite easy to do data augmentation that works.

While studying for Part1, I did an image classification where I only had 5 images per class, 102 classes. I had to train and validate with only those 5 images per class. The accuracy I got was around 85%, which positively surprised me.

Another thing I would point out is that you said that it can be cumbersome to collect a large number of pictures in real applications. One thing that is very common is to just use video and then get the frames from it.