The repository comprises of notebooks that evaluate different models for image similarity. The UI folder consist of a Tornado based application that finds the top 5 celebrity matches for a given input image.
Structure of the notebooks:
-
Data_Subsetting_train_test_split.ipynb - This notebook generates a subset of the actual LFW dataset. It also split the dataset into train and test sets for use by other notebooks.
-
Image_Similarity_Baseline_Celeb.ipynb - This notebook investigates the baseline model(VGG16)
-
Triplet_loss_celeb_cnn.ipynb - In this notebook, we implemented a custom neural network as the base architecture for triplet loss.
-
Triplet_loss_celeb_samll_cnn.ipynb - This notebook comprises of different base architectures that we evaluated.
-
Triplet_loss_celeb_resnet.ipynb - In this notebook, RESNET50 pre-trained model is used as the base model and is compiled using triplet loss function.