Context I built a Raspberry Pi Lego Sorter that used Tensorflow and Inception V3 to identify Lego Classes. I used this data set to see how far an out of the box Inception implementation would yield. You can read more about the project and sorter below: https://medium.com/@pacogarcia3/identify-lego-vs-generic-bricks-with-80-accuracy-using-vanilla-inception-v3-and-tensorflow-9e3a27cdb3ff https://medium.com/@pacogarcia3/tensorflow-on-raspbery-pi-lego-sorter-ab60019dcf32 This data set should enable to replicate my project without having to build the mechanical part. Content Contains 12 classes across 6 brick types, and more than 20,000 images taken by 4 cameras Acknowledgements See my medium post. Inspiration The journey I embarked on was inspired by a Google Post: https://cloud.google.com/blog/products/gcp/how-a-japanese-cucumber-farmer-is-using-deep-learning-and-tensorflow