Download the data from kaggle Dogs vs Cats Competition. https://www.kaggle.com/c/dogs-vs-cats-redux-kernels-edition/data
The train folder contains 25,000 images of dogs and cats. Each image in this folder has the label as part of the filename. The test folder contains 12,500 images, named according to a numeric id. For each image in the test set, you should predict a probability that the image is a dog (1 = dog, 0 = cat).
First divided the dataset into the following sub folders:
train – This is the training set used for training the model. Consists of 11500 images of cats and 11500 images of dogs.
test1 – Final test set accuracy to determine the accuracy of the model. Consists of 12500 images of both Cats and Dogs.
valid – Validation set for checking the models accuracy. Consists of 1000 cats and 1000 dogs images.
sample – Small dataset on which testing can be done so that we can save a lot of time and check the convergence of model. Consists of 12 images of each.
Cats
Dogs
Tensorflow, Python, Opencv, scikit learn.
Using Convolutonal Layers we train the model and test the model accuracy on test data. First used simple layer of 2 convolutional network and got an accuracy of 78%. Next used 8 layers Convolutional networks and got an accuracy of 92%. Next using Pretrained model VGG16 got an accuracy of 98%. Next using Pretrained model RESNET50 got an accuracy of 98.5%.
For getting State of Art results for this problem, I followed the techniques presented in the fast.ai. These are the techniques I used for getting these results.
We're going to use a pre-trained model, that is, a model created by some one else to solve a different problem. Instead of building a model from scratch to solve a similar problem, we'll use a model trained on ImageNet (1.2 million images and 1000 classes) as a starting point. The model is a Convolutional Neural Network (CNN), a type of Neural Network that builds state-of-the-art models for computer vision.
RESNET34
Architecture:
Then we start training the model by selecting a lower learning rate value.
These are the results after first training session:
[0.02893] – Training error
0.987 – Validation accuracy
A few correct labels at random
A few incorrect labels at random
The learning rate determines how quickly or how slowly you want to update the weights (or parameters). Learning rate is one of the most difficult parameters to set, because it significantly affect model performance.If we take too higher learning rate the model will not converge and if we take too smaller rate the model will take forever to converge.So finding a proper learning rate would help in saving the time spent training the model.
So we slowly increase the learning rate till our loss function starts becoming worse.We will get a graph like this
We can also plot the graph of Learning rate vs Loss
We can see the above plot of loss versus learning rate to see where our loss stops decreasing.The loss is still clearly improving at lr=1e-2 (0.01), so that's what we use here in this model.
If you try training for more epochs, you'll notice that we start to overfit, which means that our model is learning to recognize the specific images in the training set, rather than generalizing such that we also get good results on the validation set. One way to fix this is to effectively create more data, through data augmentation. This refers to randomly changing the images in ways that shouldn't impact their interpretation, such as horizontal flipping, zooming, and rotating.
Sample images after Zooming,flipping
Training Error - 0.02305 Validation accuracy - 0.989
Our validation loss isn't improving much, so there's probably no point further training the last layer on its own.So we unfreeze the layers and finetune it.
Now that we have a good final layer trained, we can try fine-tuning the other layers. To tell the learner that we want to unfreeze the remaining layers
Training loss - 0.026889
Validation Accuracy - 0.992
