This repository consist of three different types of technique that can be used for image classification task by using very little amount of dataset. These models are developed in such a way that they can distinguish between images , in this particular case i made a binary image classifier to differentiate between images of cats and dogs. The techniques that are used here are :-

• 1st_convNet.ipynb :- Training a small network from scratch (as a baseline model).
• 2nd_convNet.ipynb :- Using the bottleneck features of a pre-trained network.
• 3rd_convNet.ipynb :- Fine-tuning the top layers of a pre-trained network.

Dataset is available Over here. It is a big dataset so sample out only 1000 images of each class(dog and cat) for training and about 300 images for validation . Store these images in the following manner :-

• A directory must contain one subdirectory per class where each subdirectory will be treated as a different class, so here label of an image is set in the form of directory name.
• Images in a directory will be consider to have label as the name of that directory that they are stored in .

• For the detailed explanation from scratch , refer The Keras Blog

You can observe just by using transfer learning we can boost the accuracy of model by a huge margin , almost upto the optimal solution.