Very Deep Convolutional Networks for Large-Scale Image Recognition
Citations
421 citations
420 citations
420 citations
Cites background from "Very Deep Convolutional Networks fo..."
...Initialization is essential to train very deep networks [21, 45, 24], especially for tasks of dense prediction, where Batch Normalization [30] is less effective because of the small minibatch due to the large memory consumption of fully convolutional networks....
[...]
..., AlexNet [33] and VGGnets [45]), multi-branch networks exhibit better performance on various vision tasks....
[...]
...In contrast to traditional plain networks (e.g., AlexNet [33] and VGGnets [45]), multi-branch networks exhibit better performance on various vision tasks....
[...]
...However, it has difficulty in training very deep networks due to the instability of gradients [45]....
[...]
420 citations
Cites methods from "Very Deep Convolutional Networks fo..."
...To leverage features trained from a larger image domain, we use the sizable FCN-VGG network architecture from [18] and initialize the network weights using a model pre-trained on ImageNet for 1000-way object classification....
[...]
...More explicitly, we train a VGG architecture [18] Fully Convolutional Network (FCN) [2] to perform 2D object segmentation....
[...]
419 citations
Cites methods from "Very Deep Convolutional Networks fo..."
...For vision, we experimented with AlexNet [26], VGG [36], ResNet [18], and Inception with Batch Normalization [22] deep networks....
[...]
...However, 1BitSGD matches the performance of 2-bit and 4-bit QSGD on AlexNet, VGG, and LSTMs within 10%....
[...]
...We explore the practicality of QSGD on a variety of state-of-the-art datasets and machine learning models: we examine its performance in training networks for image classification tasks (AlexNet, Inception, ResNet, and VGG) on the ImageNet [12] and CIFAR-10 [25] datasets, as well as on LSTMs [19] for speech recognition....
[...]
...First, based on the ratio of communication to computation, we can roughly split networks into communication-intensive (AlexNet, VGG, LSTM), and computation-intensive (Inception, ResNet)....
[...]
References
49,639 citations
21,729 citations
9,803 citations
9,775 citations
6,061 citations