https://github.com/metachi/fastaiv2keras

This is an implementation of the fastai part1 v2 course in Keras
https://github.com/metachi/fastaiv2keras

Last synced: 9 days ago
JSON representation

This is an implementation of the fastai part1 v2 course in Keras

• Host: GitHub
• URL: https://github.com/metachi/fastaiv2keras
• Owner: metachi
• Created: 2017-11-09T04:02:29.000Z (about 7 years ago)
• Default Branch: master
• Last Pushed: 2018-03-09T16:05:43.000Z (over 6 years ago)
• Last Synced: 2024-08-02T15:21:17.776Z (3 months ago)
• Language: Jupyter Notebook
• Size: 424 KB
• Stars: 27
• Watchers: 4
• Forks: 3
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# fastaiv2keras

This is an implementation of the fastai part1 v2 course in Keras

lesson1 and lesson1-finetune2 jupyter notebooks go through the dogs and cats dataset

- lesson1 uses a finetune function that simply freezes early layers and makes the fc layer output predictions for 2 classes

- lesson1-finetune2 uses finetune2 function.  It uses a few extra layers to enhance the model.  It adds average and max pooling layers and concatenates them and then follows them with batchnorm, dropout, and dense layers.  Note: that our function to find the optimal learning rate (LR_FIND) does not seem to work quite as well when we use finetune2.