GoogLeNet

노트

말뭉치

1. The first version entered the field in 2014, and as the name "GoogleNet" suggests, it was developed by a team at Google.
2. The GoogleNet Architecture is 22 layers deep, with 27 pooling layers included.
3. GoogleNet is trained using distributed machine learning systems with a modest amount of model and data parallelism.
4. From the name “GoogLeNet”, we’ve already known that it is from Google.
5. To classify new images using GoogLeNet, use classify .
6. You can retrain a GoogLeNet network to perform a new task using transfer learning.
7. You can also create a trained GoogLeNet network from inside MATLAB by installing the Deep Learning Toolbox Model for GoogLeNet Network support package.
8. If the Deep Learning Toolbox Model for GoogLeNet Network support package is not installed, then the function provides a link to the required support package in the Add-On Explorer.
9. You can create an untrained GoogLeNet network from inside MATLAB by importing a trained GoogLeNet network into the Deep Network Designer App and selecting Export > Generate Code.
10. The exported code will generate an untrained network with the network architecture of GoogLeNet.
11. GoogLeNet introduces several new concepts.
12. Another key network that we already seen on quite some occasions in this lecture is GoogleNet.
13. One particular incarnation of this architecture, GoogLeNet, a 22 layers deep network, was used to assess its quality in the context of object detection and classification.
14. The paper proposes a new type of architecture – GoogLeNet or Inception v1.
15. Now that you have understood the architecture of GoogLeNet and the intuition behind it, it’s time to power up Python and implement our learnings using Keras!
16. The GoogLeNet builds on the idea that most of the activations in a deep network are either unnecessary(value of zero) or redundant because of correlations between them.
17. So GoogLeNet devised a module called inception module that approximates a sparse CNN with a normal dense construction(shown in the figure).
18. Let us take the first inception module of GoogLeNet as an example which has 192 channels as input.
19. Use of a large network width and depth allows GoogLeNet to remove the FC layers without affecting the accuracy.
20. The most straightforward way to improve performance on deep learning is to use more layers and more data, googleNet use 9 inception modules.
21. Bellow we present 2 inception layers on cascade from the original googleNet.
22. 7.4.2, GoogLeNet uses a stack of a total of 9 inception blocks and global average pooling to generate its estimates.
23. We can now implement GoogLeNet piece by piece.
24. The GoogLeNet model is computationally complex, so it is not as easy to modify the number of channels as in VGG.
25. Moreover, two popular GoogLeNet architecture versions of CNN, namely, Inception-v1 and Inception-v3, were used.
26. In GoogLeNet architecture, there is a method called global average pooling is used at the end of the network.
27. GoogLeNet was the winner at ILSRVRC 2014 taking 1st place in both classification an detection task.
28. The design of this initial Inception Module is known commonly as GoogLeNet, or Inception v1.
29. GoogLeNet uses 9 inception module and it eliminates all fully connected layers using average pooling to go from 7x7x1024 to 1x1x1024.
30. from_pretrained ( "googlenet" ) model .