This is the 2D object recognition part for the Amazon Picking Challenge using Caffe.

Directories and Contents

training

This directory contains files relavent to training of the neural network.

classifying

Files relavent to taking a pretrained neural network and image and giving its classification.

mask_formatting

All Amazon Picking Challenge images come with a mask of where the object is. This directory contains code that will use those masks and create smaller images of just the objects for training.

image_warp

All images contain white background, allowing the neural network to learn that should the background change, it's most likely due to the object being different. This is not true of real data, so this dirctory contains code that will take an image and perform various warpings on it (adding noise, changing the background, etc.) and creates more training data.