Very Deep Convolutional Networks for Large-Scale Image Recognition
Citations
1,231 citations
Cites background or methods from "Very Deep Convolutional Networks fo..."
...egularization at each layer, which is an approach that is generally uncommon in other elds such as computer vision and natural language processing. For example, the computer vision models proposed by [29] use Dropout only after dense layers, while [71] 12 suggests that using Batch Normalization may remove the need for Dropout entirely. In contrast, we found that Batch Normalization with Dropout improv...
[...]
...s managed only a few percentage points. Subsequent models published by various research groups have improved upon this approach, with current classication performance rivaling that of human labelers [27,29,30]. The main advantage of this approach lies in the ability of the network to automatically extract relevant features for the problem at hand, rather than relying on manual feature extraction approaches...
[...]
...sy to see that both approaches model the same segment of data; however, the second approach represents a signicant reduction in the total number of parameters. This parametrization trick was used by [29] as a way to eciently build deep convolutional networks for image analysis; by replacing a 7 7 kernel size with 3 layers of 3 3 kernels, one can represent the same amount of data with about a 44% red...
[...]
1,218 citations
1,218 citations
1,215 citations
Cites background from "Very Deep Convolutional Networks fo..."
...Owing to advanced network architectures [13, 23, 29, 4] and discriminative learning approaches [25, 22, 34], deep CNNs have boosted the FR performance to an unprecedent level....
[...]
1,211 citations
References
49,639 citations
21,729 citations
9,803 citations
9,775 citations
6,061 citations