This easy-to-use app proves that it is possible to run state of the art Deep Learning-based image compression algorithms on a resource-constrained device. The app features Deep Learning-based image encoder, decoder as well as an Entropy encoder to generate a comparable bitstream to traditional image compression algorithms.
The Deep Learning models used in the app are trained used InterDigital’s open source CompressAI project (https://github.com/InterDigitalInc/CompressAI) which is a Pytorch library and evaluation platform for end-to-end compression research. Pytorch versions of these models are available in the CompressAI model Zoo.
The app features a slick slider-based mechanism for comparing the visual quality of the compressed image generated from the Deep Learning based models as well as traditional image compression approaches like JPEG and HEIC.