Facebook AI Model SEER Learn To Recognize 1 Billion Instagram Images
Facebook has reached a new milestone to self-supervised learning. Its new AI model SEER was able to train itself to recognize one billion unlabeled Instagram pictures.
Facebook AI Model SEER
Facebook AI model SEER (SElf-supERvised) is based on the self-supervised learning approach. This new computer vision model can learn about any random group of images on the web. Even if the images are not labelled or annotated that typically requires for the computer vision training.
SEER was tested on one billion random and unlabeled public Instagram images and its performance was breathtaking. As its accuracy reached to 84.2 % on the ImageNet. Thus, Facebook AI team said it would be the “most advanced, state-of-the-art self-supervised systems”.
Self-supervised learning
Self-supervised learning is a new computer vision and is much appreciated by various researchers in various fields. As it not only takes less human intervention and learning, but is also a more time-consuming process. However, a self-supervised learning is more focused on self-learning, without needing accurate dataset, labelling or annotation.
Self-supervised computer vision
The blogpost explains that the SEER works parallelly to NLP, where trillions of parameters and data sets with trillions of unlabeled text used by the state-of-the-art models as pre-training. This is going to dramatically increase its performance with more datasets and ever-larger models.
Additionally, this same should stand true in computer vision.
Read full post on: https://techgeekspace.com/facebook-ai-model-seer-learn-to-recognize-1-billion-instagram-images/
