VGGNet

노트

말뭉치

1. VGGNet is a Convolutional Neural Network architecture proposed by Karen Simonyan and Andrew Zisserman from the University of Oxford in 2014.
2. Another variation of VGGNet has 19 weight layers consisting of 16 convolutional layers with 3 fully connected layers and same 5 pooling layers.
3. In both variation of VGGNet there consists of two Fully Connected layers with 4096 channels each which is followed by another fully connected layer with 1000 channels to predict 1000 labels.
4. VGGNet is a neural network that performed very well in the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) in 2014.
5. VGGNet apparently took 2-3 weeks to train on a computer with four NVIDIA Titan Black GPUs.
6. Let’s say you want to train a network such as VGGNet to recognize faces of celebrities.
7. Since the VGGNet we’re using was trained on ImageNet, it’s really good at distinguishing between different breeds of dogs, different types of fish, and so on.
8. VGGNet is invented by Visual Geometry Group (by Oxford University).
9. The reason to understand VGGNet is that many modern image classification models are built on top of this architecture.
10. It achieves better accuracy than VGGNet and GoogLeNet while being computationally more efficient than VGGNet.
11. The architecture is similar to the VGGNet consisting mostly of 3X3 filters.
12. From the VGGNet, shortcut connection as described above is inserted to form a residual network.
13. VGGNet was a competitor in the ImageNet ILSVRC-2014 image classification competition and scored second place.
14. The runner-up in ILSVRC 2014 was the network from Karen Simonyan and Andrew Zisserman that became known as the VGGNet.
15. A downside of the VGGNet is that it is more expensive to evaluate and uses a lot more memory and parameters (140M).
16. Lets break down the VGGNet in more detail as a case study.
17. The whole VGGNet is composed of CONV layers that perform 3x3 convolutions with stride 1 and pad 1, and of POOL layers that perform 2x2 max pooling with stride 2 (and no padding).
18. They compare HybridNet, with VGGNet and CBP-CNN, for 292, 100, and 200 sub-classes of VegFru Dataset.