Autoencoder for Image Retrieval

Contributors: Maximilian Lindinger & Julian Schelb

File Structure

The code of this project is divided into several jupyter notebooks loosely corresponding to the assignments sections:

Feature Encoding:

• 3_1_Feature_Encoding_Dataset_Preparation.ipynb: Importing, Splitting, Exploring and Normalizing the Image Data
• 3_2_Feature_Encoding_Build_Autoencoder.ipynb: Experimenting with different Network Architectures
• 3_3_Feature_Encoding_Train_Autoencoder.ipynb: Training using the final Network Architecture

Sanity Check & Data Querying:

• 4_Sanity_Check.ipynb: Analysing the performance of the model by comparing the original and the decoded image as well as the extracted features using a scatter plot matrix and Umap projection
• 5_Data_Querying.ipynb: Usage of the trained model to find the most similar images

Bonus:

• 6_1_2_Bonus_Retraining.ipynb: Further optimization the the model by training the model with rotated and noisy data
• 6_3_Bonus_Querying_Modified_Data: Evaluation of the optimized model

Requirements

To install the required python packages please use the following command:

pip install -r requirements.txt