NN 26 Scene Style Classification

This project contains solutions and experiments for the NN 26 Scene Style Classification Kaggle competition. The goal of the project is to classify indoor scenes into different architectural or interior design styles.

The repository includes separate Jupyter Notebooks for training, evaluating, and generating predictions using various state-of-the-art computer vision models:

• ResNet (resnet.ipynb)
• Vision Transformer (ViT) (vit.ipynb)
• ConvNeXt Base (convnext-base.ipynb)
• DINOv3 (dinov3.ipynb)

Setup & Requirements

1. Make sure you have Python installed. It is recommended to use a virtual environment.
2. Install the required dependencies from your terminal:

   pip install -r requirements.txt

Downloading the Dataset

Ensure that you have set up your Kaggle API key (a kaggle.json file placed in your ~/.kaggle/ directory).

Run the following Kaggle CLI command at the root directory of this project to download the competition dataset:

kaggle competitions download -c nn-26-scene-style-classification

Once the download finishes, unzip the file within the same root directory. The directory structure will extract naturally into the current directory.