This is a SalGAN implementation without the pre-trained VGG features written in PyTorch.
For all other information and a detailed report and metrics, see Final Report.pdf
Clone the repo. Download the trained generator weights here. Run Test_Model.ipnyb using Jupyter Notebook and use functions saliency_from_link
and saliency_from_val
Download the images and fixation maps from SALICON and save this data in a folder called data
.
Clone this repository. Install dependencies.
pip install -r requirements.txt
Rezise the images into 256x192
python resize_data.py
Train the model
python main.py