This is done as a part of Deep Learning Course undertaken during 7th Semester at Indian Institute of Space Science and Technology.
To build a model which is able to identify the source of the image dataset (i.e. the camera model).
The dataset consists of 2750 training images from 10 different camera models (classes) along with 2640 test images. Some of the test images are manipulated using different technique like JPEG compression, bicubic interpolation and gamma correction. 10 classes:
- HTC-1-M7
- iPhone-4s
- iPhone-6
- LG-Nexus-5x
- Motorola-Droid-Maxx
- Motorola-Nexus-6
- Motorola-X
- Samsung-Galaxy-Note3
- Samsung-Galaxy-S4
- Sony-NEX-7
- Cropping training images to a fixed size of 512*512 and loading the training images.
- Splitting the test images into 2 categories : Manipulated and Unaltered and loading it.
- Building a model/architecture which trains on the training dataset.
- Restoring the manipulated images.
- Combining the unaltered and restored images into the test set.
- Find the predicted labels. Link of the Kaggle competition: https://www.kaggle.com/c/sp-society-camera-model-identification
Model trained with over 90 % training as well as validation accuracy.