Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/aman5319/birdsclassification

This repo replicates all the techniques used in building a classifier in Fastai to keras. From preparing Dataset to getting state of the art result
https://github.com/aman5319/birdsclassification

createdataset fastai fastaitokeras keras

Last synced: 12 days ago
JSON representation

This repo replicates all the techniques used in building a classifier in Fastai to keras. From preparing Dataset to getting state of the art result

Awesome Lists containing this project

README 

        
# IndianBirdsClassification



Here are two important Notebook



1. [IndianBirdsClassifier.ipynb](https://github.com/aman5319/BirdsClassification/blob/master/IndianBirdsClassifier.ipynb)  

This notebook consist of Image classifier built using Fast AI library, steps performed in this notebook



	​		1. [Creating your own dataset from Google Images](https://render.githubusercontent.com/view/ipynb?commit=58c23e3dc66ca42896b1a23e776be3d59fdbd3a6&enc_url=68747470733a2f2f7261772e67697468756275736572636f6e74656e742e636f6d2f4149362d42616e67616c6f72652d436861707465722f323031382d6379636c652d322f353863323365336463363663613432383936623161323365373736626533643539666462643361362f53657373696f6e732f53657373696f6e5f31322f7068617365315f73616d706c652e6970796e62&nwo=AI6-Bangalore-Chapter%2F2018-cycle-2&path=Sessions%2FSession_12%2Fphase1_sample.ipynb&repository_id=143403708&repository_type=Repository#Creating-your-own-dataset-from-Google-Images)   



	​		2. Use transfer learning on ResNet34 model trained on Imagenet



	​		3. Use Lr find and One cycle policy method to get faster results .



	​		Finally The model accuracy is **91%**  



------

2. [Creating_Datasets.ipynb](https://github.com/aman5319/BirdsClassification/blob/master/Creating_Datasets.ipynb)  



   ​	All the above steps defined in the above IndianBirdsClassifier.ipynb are done using Fastai library, where all we do is just simple library method calls and we get a very good result



   ​	But What this repo focuses on replicating all of those into Keras



   ​	So This Notebook shows  how to easily create an image dataset through Google Images and load them in to keras to train your model .

_____



Birds.zip file contains 10 CSV files of different birds. With each CSV file having the image URL of the birds.



3. [LrFinder.py](https://github.com/aman5319/BirdsClassification/blob/master/LrFinder.py) 



   ​	This class uses the Cyclic Learning Rate history to find a set of learning rates that can be good  initializations for the One-Cycle training proposed by Leslie Smith in the paper referenced  below. 

   

   ​	A port of the Fast.ai implementation for Keras.

   

   ​	Interpretation

   

   ​	Upon visualizing the loss plot, check where the loss starts to increase rapidly. Choose a learning rate at somewhat prior to the corresponding position in the plot for faster convergence. This will be the max_lr.

   

   ​        References:

   ​            [A disciplined approach to neural network hyper-parameters: Part 1 -- learning rate, batch size, weight_decay, and weight decay](https://arxiv.org/abs/1803.09820)



   ```python

   #usage

   from LrFinder import LRFinder

   

   lrfind = LRFinder(max_iteration = len(feature_train)//batch_size )

   history = classifier.fit(feature_train,

                            label_train,

                            epochs=1,

                            batch_size=batch_size,

                            shuffle=True ,

                            callbacks=[lrfind] )

   

   ```



4. [OneCyclePolicy.py](https://github.com/aman5319/BirdsClassification/blob/master/OneCyclePolicy.py)



   ​	This callback implements a cyclical learning rate policy (CLR). This is a special case of Cyclic Learning Rates, where we have only 1 cycle. After the completion of 1 cycle, the learning rate will decrease rapidly to 10000th its initial lowest value.

      Reference



   ​	[A disciplined approach to neural network hyper-parameters: Part 1 -- learning rate, batch size, weight_decay, and weight decay](https://arxiv.org/abs/1803.09820)

   ​	[Super-Convergence: Very Fast Training of Residual Networks Using Large Learning Rates](https://arxiv.org/abs/1708.07120)



   ```python

   #usage

   from OneCyclePolicy import OneCycleScheduler

   

   fit_one_cycle = OneCycleScheduler(num_iteration = len(feature_train)//batch_size ,

                                     num_epochs =4 ,

                                     max_lr = 2e-3) #max_lr is the lr you got from  lrfind

   classifier.fit(feature_train,

                  label_train ,

                  epochs=4,

                  batch_size=batch_size,

                  shuffle=True,

                  callbacks=[fit_one_cycle],

                  validation_data=(feature_test,label_test))

   

   ```



The [Mnist_callbacks_test.ipynb](https://github.com/aman5319/BirdsClassification/blob/master/Mnist_callbacks_test.ipynb) file contains the test of above two callbacks