GrokNet
Authors:
Open Access
- 20 August 2020
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 2608-2616
Abstract
In this paper, we present GrokNet, a deployed image recognition system for commerce applications. GrokNet leverages a multi-task learning approach to train a single computer vision trunk. We achieve a 2.1x improvement in exact product match accuracy when compared to the previous state-of-the-art Facebook product recognition system. We achieve this by training on 7 datasets across several commerce verticals, using 80 categorical loss functions and 3 embedding losses. We share our experience of combining diverse sources with wide-ranging label semantics and image statistics, including learning from human annotations, user-generated tags, and noisy search engine interaction data. GrokNet has demonstrated gains in production applications and operates at Facebook scale.Keywords
This publication has 13 references indexed in Scilit:
- Exploring the Limits of Weakly Supervised Pretraining2018
- Visual Search at Alibaba2018
- Fine-Tuning CNN Image Retrieval with No Human AnnotationIEEE Transactions on Pattern Analysis and Machine Intelligence, 2018
- NormFace2017
- Revisiting Unreasonable Effectiveness of Data in Deep Learning Era2017
- Visual Search at eBay2017
- Learning Visual Features from Large Weakly Supervised Data2016
- Cross-Stitch Networks for Multi-task Learning2016
- Learning visual similarity for product design with convolutional neural networksACM Transactions on Graphics, 2015
- SIGNATURE VERIFICATION USING A “SIAMESE” TIME DELAY NEURAL NETWORK1994