Trying to figure out who is in that celebrity GIF? This AI-based algorithm can help. Researchers from GIPHY, the online search database for GIFs, recently developed an open source deep learning model that can recognize over 2,300 celebrity faces with high accuracy.
“We needed a tool that could find and annotate this content within our ever-growing library of GIFs, so that this content could then be found in our search engine,” said Nick Hasty, Director of R&D at GIPHY.
Using NVIDIA Tesla V100 GPUs, with the cuDNN-accelerated TensorFlow deep learning framework, the researchers trained a convolutional neural network, based on ResNet 50, to perform face detection and face recognition on thousands of celebrities like 50 Cent, Adam Sandler, George Clooney, Halle Berry, Will Ferrell and many more.
“When a GIF or image is submitted to the classifier, it attempts to detect all faces across all frames using a popular pre-trained model,” said Hasty. “An aggregate prediction is computed for all the faces within the cluster yielding one or more celebrity names, each with a confidence score. The model’s final output is the combination of these predictions across all clusters.”
To validate the results the team tested the classifier on crowdsourced labeled and verified images comprised of over 1,000 popular GIPHY celebrities. On that dataset, the system achieved a 98% precision score.
Hasty and his team say that over the coming months they will describe how they tested the model for different types of bias.
The company hopes that developers will be inspired by the code, replicate it, and perhaps adapt it into their own work.
Anyone can try the model on the company’s demo page and download the code via GIPHY’s Github page, as well as see a 3D projection of the work.
