Deep Image: Scaling up Image Recognition

By Ren Wu, Shengen Yan, Yi Shan, Qingqing Dang, Gang Sun. 2015

385 citations recorded by the source corpus.

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.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift (2015, neutral): “The current reported best results on the ImageNet Large Scale Visual Recognition Competition are reached by the Deep Image ensemble of traditional models (Wu et al., 2015) and the ensemble model of (He et al., 2015).”
Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification (2015, background): “This is also a ∼ 17% relative improvement over the latest result (Baidu, 5.98% [32]).”
LSUN: Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop (2015, background): “High capacity supervised learning algorithms, such as deep convolutional neural networks [10, 9], have led to a discontinuity in visual recognition over the past four years and continue to push the state-of-the-art performance (e.g. [7, 22, 8]).”
Improved Regularization of Convolutional Neural Networks with Cutout (2017, background): “take a more aggressive approach with image augmentation when training Deep Image [21] on the ImageNet dataset.”
ScaleNet: An Unsupervised Representation Learning Method for Limited Information (2023, background): “Multiple studies employed data augmentation techniques such as scaling, rotating, cropping, and generating synthetic samples to produce more training samples [22,45,23].”
Deep Learning with Limited Numerical Precision (2015, methodology): “or a combination of CPUs and GPUs scaled-up to multiple nodes (Coates et al., 2013; Wu et al., 2015).”
Action Recognition using Visual Attention (2015, background): “Convolutional Neural Networks (CNNs) have been highly successful in image classiﬁcation and object recognition tasks (Ren et al., 2015; Wu et al., 2015).”
Probabilistic Backpropagation for Scalable Learning of Bayesian Neural Networks (2015, background): “…linear units (Nair & Hinton, 2010), have resulted in successes for NNs on tasks such as speech recognition (Hinton et al., 2012; Hannun et al., 2014), computer vision (Krizhevsky et al., 2012; Wu et al., 2015) and natural language processing (Collobert & Weston, 2008; Sutskever et al., 2014).”
APAC: Augmented PAttern Classification with Neural Networks (2015, background): “More recently, Wu et al . vastly expanded the ImageNet dataset with many types of image processing including color casting, vignetting, rotation, aspect ratio change, and lens distortion on top of standard cropping and ﬂipping [22].”
FireCaffe: Near-Linear Acceleration of Deep Neural Network Training on Compute Clusters (2015, methodology): “Amazon and Baidu2 each reported their training time numbers on a proprietary dataset, so it’s not clear how to compare these results with the related literature.”
CaffePresso: An optimized library for Deep Learning on embedded accelerator-based platforms (2016, background): “The low efficiency of the GPU for small ConvNets is primarily due to the CUDA launch overheads, and lack of low-level control of memory transfers (CUDA-compatible GPUs do allow cache prefetch in __shared__ 64 KB memory).”
Data Augmentation in Classification and Segmentation: A Survey and New Strategies (2023, methodology): “Colour jittering is a photometric data augmentation technique that employs either random colour manipulation [42] or predetermined colour adjustments [43], such as randomly changing the brightness, contrast or colour properties of an image, see Figure 8.”
Construction of a Large-scale Image Dataset using Deep Learning with Humans in the Loop (2015, methodology): “[17]) to train these models, there is little progress on pushing forward the data construction efforts to the next level.”
YodaNN: An Ultra-Low Power Convolutional Neural Network Accelerator Based on Binary Weights (2016, background): “Convolutional Neural Networks (CNN) algorithms have been achieving outstanding classiﬁcation capabilities in several complex tasks such as image recognition [2], [3], face recognition [4] and speech recognition [5].”
Fine-Grained Image Classification by Exploring Bipartite-Graph Labels (2015, background): “While general image classification has achieved impressive success within the last few years [31, 49, 46, 21, 56, 22], it is still very challenging to recognize object classes with ultra-fine granularity.”