Reducing The Amortization Gap of Entropy Bottleneck In End-to-End Image Compression

End-to-end trainable models are about to exceed the performance of the traditional handcrafted compression techniques on videos and images. The core idea is to learn a non- linear transformation into latent space, jointly with an entropy model of the latent distribution. These methods enforce the latent to follow some prior distributions. Since the priors are learned by amortizing its parameters over the entire training set, it cannot fit exactly on every single instance. As a consequence, it damages the compression performance by enlarging the bitstream. In this paper, we proposed a simple yet efficient instance-based parametrization method to reduce this amortization gap at a minor cost. The proposed method is applicable to any end-to- end compressing methods, improving the compression bitrate by 1% without any impact on the reconstruction quality.