Autoencoder based image compression: can the learning be quantization independent?
This addresses the efficiency issue in autoencoder-based image compression for researchers and practitioners by reducing the need for multiple trained models.
The paper tackles the problem of learning image compression transforms with autoencoders, showing that a single learned transform can achieve comparable rate-distortion performance across different quantization step sizes, saving significant training time.
This paper explores the problem of learning transforms for image compression via autoencoders. Usually, the rate-distortion performances of image compression are tuned by varying the quantization step size. In the case of autoen-coders, this in principle would require learning one transform per rate-distortion point at a given quantization step size. Here, we show that comparable performances can be obtained with a unique learned transform. The different rate-distortion points are then reached by varying the quantization step size at test time. This approach saves a lot of training time.