The image dataset is a crossroad between human vision and machine vision. Humans developed this dataset to enable machines to understand the world through our perspective, effectively creating a prosthetic eye that offers inexhaustible labor to serve us. But does this mean machines will see the world as we do?

This project investigates the new regime of representation constructed by machine vision. Gravitated towards the blackbox of deep learning and its appetite for data, I discussed about the automation paradox within our visual culture. The project presents 110 selected training images from sixteen food dish classes in the ImageNet-1k database. Labels with scores are interpreted by Google Cloud Vision, and sixteen neural network output images corresponding to each class are generated by the ResNet50V2 model.

Special thanks to Yiming Yan for coding support.