Efficient Learned Image Compression without Entropy Coding
Researchers have developed a new method for learned image compression called EF-LIC, which eliminates the need for traditional entropy coding. This approach significantly reduces coding latency by removing statistical and correlation redundancy through unconstrained vector quantization and a context-conditioned autoregressive transform. Experiments demonstrate that EF-LIC achieves comparable compression performance to existing methods while offering substantial speed improvements, with over 3x faster encoding and 5x faster decoding. AI
IMPACT Introduces a novel technique for image compression that significantly speeds up encoding and decoding processes.