This project consists of the implementation of several hierarchical classification models. The goal is to classify images into a hierarchical structure, where each class has a parent class. The dataset used is the ImageNet dataset, which consists of 1000 classes. The hierarchical structure has been extracted from WordNet.

We implemented the following models:

• VGG16​: based on the original paper
• HVGG16​ and HVGG11: based on the original paper
• ResNet50​: based on the original paper
• CondResNet50: based on the original paper
• HResNet​50: based on ResNet50, but with hierarchical classification as in HVGG16
• Branch Selection ResNet​50: based on ResNet50, but with branch selection mechanism based on the coarse prediction
• CondHResNet50: based on HResNet50, but with conditionally parameterized convolutions

Here we can see the validation accuracy over training time (50 epochs) of the ResNet50-based models:

Install the conda environment using the environment.yml file:

conda env create -f environment.yml

Download the data from here and extract it to the dataset/ folder.

Now you can run the val2train_mapping.py to correctly organize the validation data:

python -m moduels.utils.val2train_mapping.py

Now you should have the following folder structure:

dataset/
    train/
        n01440764/ # e.g. synset
            n01440764_10026.JPEG # e.g. image
            ...
    val/
        n01440765/ # e.g. synset
            n01440765_10026.JPEG # e.g. image
            ...
    test/
        n01440766/ # e.g. synset
            n01440766_10026.JPEG # e.g. image
            ...
    hierarchy.csv
    val_mapping.csv

To train the model, run the run.py file:

python run.py