CompareAI 1.0

This application demonstrates that advanced Deep Learning-based image compression algorithms can be effectively implemented on resource-constrained devices. It includes encoders and decoders powered by deep learning techniques, complemented by an entropy encoder to produce bitstreams comparable to those generated by traditional compression methods.

The underlying models are trained using InterDigital’s open-source CompressAI platform (https://github.com/InterDigitalInc/CompressAI), which provides a PyTorch-based framework for end-to-end compression research. Corresponding PyTorch models are accessible through the CompressAI model zoo.

The user interface incorporates a smooth slider mechanism that allows for direct comparison of the visual quality between images compressed with deep learning models and those processed via conventional algorithms such as JPEG and HEIC. This facilitates an insightful evaluation of the compression performance and image fidelity across different approaches.