MIT-Adobe-FiveK-Pytorch

A Pytorch dataset class implementation for MIT-Adobe FiveK dataset.

Please refer to the dataset website to get the license and other information about the dataset: https://data.csail.mit.edu/graphics/fivek/

The code consists of:
preprocess_5k.py : Code to convert the ".dng" files into ".jpg".
MITAdobe5K.py : Dataset implementation.

Note:

1. Download the dataset here https://data.csail.mit.edu/graphics/fivek/fivek_dataset.tar and extract the dataset. The extracted dataset is assumed in a folder named "MIT-Adobe-5K".
2. Prepare a folder named "5K-jpgs". This folder should contain subdirectories named "train", "val", and "test". The file preprocess_5k.py will split, process, and save the ".jpg" images to these directories. If you want to change directory name or location, please change the variable DATASET_DIR and TARGET_DIR in the file.
3. The file MITAdobe5K.py contains the Pytorch class implementation for the dataset. Please adjust the DATASET_DIR in accordance to the preprocess_5k.py (if you make any change).