Coarse-to-Fine Hyper-Prior Modeling for Learned Image Compressionn

Training code is now available in the PyTorch repo.

A PyTorch implementation with acceleration improvements is now available!

Check this repo for details.

Overview

This is the implementation of ther paper,

Yueyu Hu, Wenhan Yang, Jiaying Liu, Coarse-to-Fine Hyper-Prior Modeling for Learned Image Compression, AAAI Conference on Artificial Intelligence (AAAI), 2020

The currently available code is for evaluation, while it can also be modified for training as the implementation of the network is available.

Running

The code requires the TensorFlow library (v1.13, v1.14 and v1.15 tested). It should be running in the CPU-only mode, for example, by specifying CUDA_VISIBLE_DEVICES= . An example to run the encoder and decoder is provided below.

You may first download the trained weights from Google Drive and place the .pk files under the models folder (that is, to make './models/model0_qp1.pk exist).

Help

python AppEncDec.py -h

Encoder

python AppEncDec.py compress example.png example.bin --qp 1 --model_type 0

Decoder

python AppEncDec.py decompress example.bin example_dec.png

Detailed command line options are documented in the help mode of the APP.