Improving the image classification using CNN, Data Augmentation, Transfer Learning and Ensemble Techniques
The objective of this assessment is to build a range of deep learning models using convolutional neural networks. The data we will be using a modified version of the Flowers 17 dataset. You can find the original dataset here. This is a multi-class classification problem with 17 possible classes (17 different classes of flowers). The images have significant variation in pose and lighting and there is also significant intra-class variation as well as close similarity between other distinct classes. What makes this dataset even more challenging is that there are only 80 images for each class. Therefore, in total there are just 1360 images.
Explores the application of convolutional networks, data augmentation and ensemble techniques.
Focuses on transfer learning (specifically the use of CNNs as feature extractors and the application of fine tuning with pre-trained CNNs).