In plain words: They built a dedicated supercomputer and a faster way to train much larger image-recognition networks, feeding them varied, high-resolution pictures at several scales. The system scored best on several hard vision tests, outperforming the usual smaller, lower-resolution setups.
Abstract · Deep Image: Scaling up Image Recognition
We present a state-of-the-art image recognition system, Deep Image, developed using end-to-end deep learning. The key components are a custom-built supercomputer dedicated to deep learning, a highly optimized parallel algorithm using new strategies for data partitioning and communication, larger deep neural network models, novel data augmentation approaches, and usage of multi-scale high-resolution images. Our method achieves excellent results on multiple challenging computer vision benchmarks.
Ren Wu, Shengen Yan, Yi Shan, Qingqing Dang, Gang Sun
arXiv:1501.02876 · cs.CV · submitted Jan 13, 2015 · updated Jul 6, 2015
abstract · pdf · This paper has been withdrawn by the authors due to a mistake related to ImageNet server submissions
