RAITE: ND Blue Team 1

This repository contains Notre Dame Blue Team's solutions for RAITE 2023. It provides scripts for training and testing two models, droid and ngebm, to classify images based on the presence of humans. These models are trained to detect whether an image contains a human or not.

Prerequisites

• Anaconda or Miniconda

Setup

1. Clone the Repository:

   git clone https://github.com/nd-crane/nd-blueteam-1.git
   cd nd-blueteam-1

2. Set Up the Anaconda Environment: Create a new Anaconda environment using the provided environment.yml file:

   conda env create -f environment.yml

   Activate the environment:

   conda activate nd-blueteam-1

3. Download the Models: The models (droid and ngebm) can be downloaded from the provided links.

   • Download Droid Model Here
   • Download Ngebm Model Here

Training

Droid Model

To train the droid model, use the following command:

python training_code/train_droid_alphas.py -datasetPath [PATH_TO_DATASET] \
    -outputPath [PATH_TO_SAVE_MODEL] \
    -network densenet \
    -alpha_xent 0.5 \
    -alpha_droid 0.5 \
    -nEpochs 50

Ngebm Model

To train the ngebm model, use the following command:

python training_code/train_droid_alphas.py -datasetPath [PATH_TO_DATASET] \
    -outputPath [PATH_TO_SAVE_MODEL] \ 
    -network densenet \
    -alpha_xent 0.5 \
    -alpha_energy_derivative 0.5 \
    -nEpochs 50

Testing

To test an image using a specific model, use the following command:

python testing_code/test_lightweight.py \
    -modelPath [PATH_TO_YOUR_MODEL]
    -imagePath [PATH_TO_YOUR_IMAGE] \

Example:

python testing_code/test_lightweight.py \
    -modelPath ../droid_model.pth \
    -imagePath mixed_singlefile_pair_standing_split1_0058.png \

Results Interpretation

The script will output a single softmax score indicating the presence of people in the image:

• If the value is over 0.5, there are no people.
• If the value is under 0.5, there are people.