2% Of Spaces You Could Possibly Encounter
'2% Of Spaces You Could Possibly Encounter' uses machine learning to classify an installation in real-time as visitors modify the layout of the space. Using MIT's Places database of 10 million images, a live feed is taken from the camera, ran through a neural network trained on the dataset, then filed into one of 365 spacial categories, ranging from amusement parks to operating theatres
Technologies used:
Python
Docker
Raspberry Pi
Responding to the claim that the Places dataset can account for 98% of spaces a user could possibly encounter, visitors are encouraged to interact with the camera feed in order to reveal the quirks, blind-spots and biases that can be found within the dataset. Results of the current classification are announced via a text-to-speech system and can also be seen on a computer monitor contained within the space.
This project draws from ongoing research into datasets; who compiles them, how they are compiled, and what they are (and can potentially be) used for.