I am a student in Serge Belongie's vision group at UCSD.

To say hello, send an email to gvanhorn {at} ucsd {dot} edu.


Currently I am interested in making expert visual knowledge accessible to mere mortals. For example, most of us know a bird when we see one, however relatively few of us can identify the species. In order to train a computer to perform an expert recognition task we need to collaborate with experts in the domain and coordinate data collection and annotation efforts. The projects that I currently work on are aimed at algorithmic solutions to the recognition task as well as minimizing the data management overhead.


A large scale, cloud based, collaborative data management system for computer vision projects.
Visipedia, short for “Visual Encyclopedia,” is an augmented version of Wikipedia, where pictures are first-class citizens alongside text.
An ongoing research project for word recognition in unconstrained images.


Bird Similarity Embedding

Bird Similarity

This is an embedding of Mturkers' perceptual similarity of bird species in the CUB-200 dataset.

Bootstrapping Classifiers>
					<div class=

Bootstrapping Classifiers

What if you had an expert give you a few positive examples of a visual event, and a huge dataset of images. Can you use the crowd to quickly build classifiers that can identify the visual event? From work with Genevieve Patterson.

Bound Box Stats Bound Box Stats

Annotation Muscle of Citizen Scientists

Visualizing the annotation power of Cornell's Lab of Ornithology citizen scientists on bounding box and part tasks. If you plan accordingly, you can collect a lot of annotations very quickly. Those spikes are "social media blasts."

Bound Box Stats

NABirds 700 Dataset

You can interact with it here, thanks to D3.