Utility Scripts and Procedures for Preparing Dataset for Learned Image Compression
Dataset used in the paper Learned Image Compression with Mixed Transformer-CNN Architectures
Official Repository: https://github.com/jmliu206/LIC_TCM
Download Imagenet repository and unzip.
Imagenet: https://www.image-net.org/download.php (130GB)
Then, randomly choose 300k images. However, some images in Imagenet are less than size 256 and not suitable for training.
So my script chooses the largest 300k images deterministicly.
Here root_dir denotes the directory where the folder imagenet-object-localization-challenge is located.
For example, if your directory path looks like /home/datasets/imagenet-object-localization-challenge/ILSVRC/Data/..., then specify /home/datasets.
The output directory is also made under /home/datasets.
python3 imagenet.py --root_dir /path/to/directory/
Dataset used in the paper MLIC ++ : Linear Complexity Multi-Reference Entropy Modeling for Learned Image Compression
Official Repository: https://github.com/JiangWeibeta/MLIC
First, download each dataset and put them in some directory.
Imagenet: https://www.image-net.org/download.php (130GB)
COCO: https://cocodataset.org/#download (81GB)
- Download ALL images; 2014 Train, 2014 Val, 2014 Test, 2015 Test, 2017 Train, 2017 Val, 2017 Test and 2017 Unlabeled.
Flickr 2k: https://www.kaggle.com/datasets/daehoyang/flickr2k (11GB)
- Note that this is not Flickr30k dataset or Flickr8k dataset!
CLIC: https://clic.compression.cc/2021/tasks/index.html
DIV2k: https://data.vision.ee.ethz.ch/cvl/DIV2K/
- Download two files labeled High Resolution Images: Train Data (HR images), Validation Data (HR images)
You can download train image list from (MLIC repo)[https://github.com/JiangWeibeta/MLIC]
Then, run select.py in this repository.
python3 select.py --train_list_path /path/to/mlic/train/list --source_dir /path/to/directory/where/all/datas/stored \
--dest_dir /path/to/directory/where/output/will/be/stored