Image Classification using Transfer Learning

Used pretrained CNNs and custom CNNs to classify dog images in project 1 and dog breeds in project 2. Transfer learning is a popular method in computer vision because it allows us to build accurate models in a timesaving way. These projects were in my Udacity's Nanodegree courses.

Transfer Learning Notebook

Notebook exploring torchvision pretrained models for image classification. View here

Project 1

CNNs used :

• VGG16
• AlexNet
• ResNet

Usage

Run following scripts:

1. First change the directory and add image in the uploaded_images folder if you want,
2. Run the following script

./run_models_batch_uploaded.sh

If the permission denied error shows up run the following before step 2

chmod u+r+x run_models_batch_uploaded.sh

Images

Project 2

In this prohect objective is to classify dog breeds using pretrained network such as ResNet and custom network architecture for comparable study. Also a exploration using HarCascade from OpenCV for face detection is performed.

Dependencies:

• Python 3.6+
• Pytorch (version 0.4+)
• OpenCV
• Numpy

Future Works:

• Increasing Accuracy
• Using more networks

References:

Udacity Deep Learning Nano Degree