UberNet demo
This website demonstrates a deep learning system trained to address a diverse set of computer vision tasks, spanning low- mid- and high- level vision.Technical details can be found in the following report:
Iasonas Kokkinos, UberNet : Training a ‘Universal’ Convolutional Neural Network for Low-, Mid-, and High-Level Vision using Diverse Datasets and Limited Memory, arxiv, 2016
Try it out here!
Please note:
- the algorithm requires less than a second per image, but the server interface is currently quite slow; images larger than 1MB may never make it to the server!
- the high-level tasks (top row) can only handle objects belonging to the 20 classes listed below. The low- and mid-level tasks (bottom row) are class-agnostic.
Indicative results:
