Parallel Neural Local Lossless Compression
This work addresses runtime efficiency for users of neural compression methods, but it is incremental as it builds on existing local autoregressive models.
The paper tackled the problem of slow runtime in neural lossless image compression by proposing two parallelization schemes for local autoregressive models, achieving significant gains in compression runtime compared to the previous non-parallel implementation.
The recently proposed Neural Local Lossless Compression (NeLLoC), which is based on a local autoregressive model, has achieved state-of-the-art (SOTA) out-of-distribution (OOD) generalization performance in the image compression task. In addition to the encouragement of OOD generalization, the local model also allows parallel inference in the decoding stage. In this paper, we propose two parallelization schemes for local autoregressive models. We discuss the practicalities of implementing the schemes and provide experimental evidence of significant gains in compression runtime compared to the previous, non-parallel implementation.