Variable-Rate Learned Image Compression with Multi-Objective Optimization and Quantization-Reconstruction Offsets

TL;DR

The paper tackles the problem of achieving variable bitrate in end-to-end learned image compression using a single neural network. It introduces three post-training modifications—multi-objective optimization (MGDA-based), quantization-reconstruction offsets, and variable-rate quantization for hyper latents—to enable continuous bitrate control without retraining multiple models. Applied to three standard models, the approach yields RD performance close to that of separately trained models, with improvements most notable at lower bitrates and minor overhead for hyper-latent quantization. This work offers a practical, post-training path to deploy flexible, variable-rate learned compression in real-world pipelines, supported by publicly available code.

Abstract

Achieving successful variable bitrate compression with computationally simple algorithms from a single end-to-end learned image or video compression model remains a challenge. Many approaches have been proposed, including conditional auto-encoders, channel-adaptive gains for the latent tensor or uniformly quantizing all elements of the latent tensor. This paper follows the traditional approach to vary a single quantization step size to perform uniform quantization of all latent tensor elements. However, three modifications are proposed to improve the variable rate compression performance. First, multi objective optimization is used for (post) training. Second, a quantization-reconstruction offset is introduced into the quantization operation. Third, variable rate quantization is used also for the hyper latent. All these modifications can be made on a pre-trained single-rate compression model by performing post training. The algorithms are implemented into three well-known image compression models and the achieved variable rate compression results indicate negligible or minimal compression performance loss compared to training multiple models. (Codes will be shared at https://github.com/InterDigitalInc/CompressAI)

Figures (3)

• Figure 1: Mean&Scale Hyperprior based image compression model minnen2018joint.
• Figure 2: Quantization with learned quantization-reconstruction offsets
• Figure 3: Comparison of compression performance on the Kodak dataset.