Improving The Reconstruction Quality by Overfitted Decoder Bias in Neural Image Compression
This incremental improvement addresses compression quality for users of neural image compression systems, with minor trade-offs in encoding time and signaling cost.
The authors tackled the problem of suboptimal neural image compression for individual images by proposing instance-based fine-tuning of decoder biases, which improved reconstruction quality with a 3-5% BD-rate gain over state-of-the-art methods.
End-to-end trainable models have reached the performance of traditional handcrafted compression techniques on videos and images. Since the parameters of these models are learned over large training sets, they are not optimal for any given image to be compressed. In this paper, we propose an instance-based fine-tuning of a subset of decoder's bias to improve the reconstruction quality in exchange for extra encoding time and minor additional signaling cost. The proposed method is applicable to any end-to-end compression methods, improving the state-of-the-art neural image compression BD-rate by $3-5\%$.