Object Shadow Cues

Getting Started

The main packages are listed below

#Conda
python=3.11.4
torchaudio=2.0.2=py311_cu117
torchvision=0.15.2=py311_cu117
tqdm=4.65.0
pillow=10.2.0
#pip
pandas==2.1.1
scikit-learn==1.3.2
matplotlib==3.8.0

Download the Trained Model here and place them in the checkpoints folder

Dataset Download

The dataset can be found here

To get started, download the Kandinsky_Indoor_OS.zip and Kandinsky_Outdoor_OS.zip

cd ../dataset
curl -L -o ./Kandinsky_Indoor_OS.zip <link to file>
curl -L -o ./Kandinsky_Outdoor_OS.zip <link to file>

where <link to file> can be found in the dataset link above and will look something like this: https://huggingface.co/datasets/amitabh3/Projective-Geometry-OS/resolve/main/Kandinsky_Indoor_OS.zip?download=true

To unzip the downloaded files:

unzip Kandinsky_Indoor_OS.zip
mv Kandinsky_Indoor_OS/* ./

To extracting Shadow and Object masks for new images that are not in the dataset, use SSISv2

Usage

Training

Run one of the following:

python train.py --category indoor
python train.py --category outdoor
python train.py --category combined

Testing

Run one of the following:

python test.py --category indoor
python test.py --category outdoor
python test.py --category combined

After testing, results will be printed and plots will be generated in the plots directory