Very Deep Convolutional Networks for Large-Scale Image Recognition
By Karen Simonyan, Andrew Zisserman. International Conference on Learning Representations, 2014
112,917 citations recorded by the source corpus.
In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.
Subfields: Computer Vision and Pattern Recognition, Media Technology, Model Architectures
View Paper: https://arxiv.org/abs/1409.1556
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
- Luc Van Gool
- Dacheng Tao
- Ming-Hsuan Yang
- Xiaogang Wang
- Trevor Darrell
- Jitendra Malik
- Yi Yang
- Andrew Zisserman
In Media Technology
- Luc Van Gool
- Radu Timofte
- Ming-Hsuan Yang
- Lei Zhang
- Chao Dong
- Wangmeng Zuo
- Kai Zhang
- Chen Change Loy
In Model Architectures
- Quoc V. Le
- Yoshua Bengio
- Alan L. Yuille
- Kaiming He
- Jiashi Feng
- Dacheng Tao
- Ruslan Salakhutdinov
- Jianfeng Gao
Citations in the corpus
- Deep Residual Learning for Image Recognition (2015, methodology): “On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [41] but still having lower complexity.”
- Factorization tricks for LSTM networks (2017, methodology): “The idea of factorization of large convolutinal layer into the stack of layers with smaller filters was used, for example, in VGG networks (Simonyan & Zisserman, 2014), and in ResNet “bottleneck design” (He et al., 2016).”
- Layer Normalization (2016, methodology): “Images and sentences from the Microsoft COCO dataset [Lin et al., 2014] are embedded into a common vector space, where a GRU [Cho et al., 2014] is used to encode sentences and the outputs of a pre-trained VGG ConvNet [Simonyan and Zisserman, 2015] (10-crop) are used to encode images.”
- Rethinking the Inception Architecture for Computer Vision (2015, background): “VGGNet [18] and GoogLeNet [20] yielded similarly high performance in the 2014 ILSVRC [16] classification challenge.”
- Learned in Translation: Contextualized Word Vectors (2017, methodology): “Qi et al. [2016] propose a CNN-based object tracking framework, which uses hierarchical features from a pretrained CNN (VGG-19 by Simonyan and Zisserman [2014]).”
- Skip-Thought Vectors (2015, methodology): “For our experiments, we represent images using 4096-dimensional OxfordNet features from their 19-layer model [35].”
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (2015, methodology): “trained network, we use the “fast” version of ZF net [32] that has five convolutional layers and three fully-connected layers, and the public VGG-16 model(8) [3] that has 13 convolutional layers and three fully-connected layers.”
- Object Detection via a Multi-region and Semantic Segmentation-Aware CNN Model (2015, methodology): “On the one hand, we build a rich candidate box representation that captures several different aspects of an object such as its pure appearance characteristics, the joint appearance on both sides of the object borders, the distinct appearance of its different regions, context appearance, and…”
- Highway Networks (2015, background): “…image classification accuracy on the 1000-class ImageNet dataset has increased from ∼ 84% (Krizhevsky et al., 2012) to ∼ 95% (Szegedy et al., 2014; Simonyan & Zisserman, 2014) through the use of ensembles of deeper architectures and smaller receptive fields (Ciresan et al., 2011a;b; 2012) in just…”
- Fast R-CNN (2015, background): “With very deep networks, such as VGG16 [19], this process takes 2.5 GPU days for the 5k images of the VOC07 trainval set.”
- Object Detection Networks on Convolutional Feature Maps (2015, methodology): “This strategy 50 was, however, historically driven by pre-trained clas- 51 sification architectures similar to AlexNet [5] and 52 VGG nets [15] that end with MLP classifiers.”
- Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification (2015, methodology): “In our four-GPU implementation, our model A takes 2.6s per mini-batch (128), and our reproduced VGG-19 takes 3.0s, evaluated on four Nvidia K20 GPUs.”
- FitNets: Hints for Thin Deep Nets (2014, background): “…are exponentially more expressive than shallow ones for some families of functions (Montufar et al., 2014); and (2) empirically: the two top-performers of ImageNet use deep convolutional networks with 19 and 22 layers, respectively (Simonyan & Zisserman, 2014) and (Szegedy et al., 2014).”
- Convolutional neural networks at constrained time cost (2014, methodology): “No other data augmentation, such as scale jittering [10, 22, 9], is used.”
- Deeply-Supervised Nets (2014, methodology): “Therefore, we have to adopt a pre-training strategy suggested in [23]: we first pre-train a relatively shallower 8-layer network and then restart the training process with another three convolutional layers for the 11-layer CNN configuration.”
- ImageNet Large Scale Visual Recognition Challenge (2014, background): “3 – – – University of Oxford Karen Simonyan, Andrew Zisserman (Simonyan and Zisserman 2014) XYZ 11.”