🤖 AI Summary
To address the demand for high-energy-efficiency, low-resource hardware accelerators for edge-based image classification and compression, this work proposes a reconfigurable ASIC neural network encoder consuming only 1 MB of on-chip resources and supporting mixed-precision computation (3-bit/2-bit/1-bit). Methodologically, it introduces a novel linear symmetric quantization with adaptive scaling factors to ensure training stability under ultra-low-bit weight quantization, and replaces batch normalization with layer-shared shift-based normalization to drastically reduce hardware overhead. Furthermore, it unifies classification and compression acceleration via structural pruning, block-wise encoding, and remote full-frame decoding. Evaluated on CIFAR-10, the encoder achieves 87.5% classification accuracy and—uniquely among learned image codecs—enables end-to-end, block-artifact-free image compression. Its fixed-bitrate performance significantly surpasses state-of-the-art block-based compression methods.
📝 Abstract
Even if Application-Specific Integrated Circuits (ASIC) have proven to be a relevant choice for integrating inference at the edge, they are often limited in terms of applicability. In this paper, we demonstrate that an ASIC neural network accelerator dedicated to image processing can be applied to multiple tasks of different levels: image classification and compression, while requiring a very limited hardware. The key component is a reconfigurable, mixed-precision (3b/2b/1b) encoder that takes advantage of proper weight and activation quantizations combined with convolutional layer structural pruning to lower hardware-related constraints (memory and computing). We introduce an automatic adaptation of linear symmetric quantizer scaling factors to perform quantized levels equalization, aiming at stabilizing quinary and ternary weights training. In addition, a proposed layer-shared Bit-Shift Normalization significantly simplifies the implementation of the hardware-expensive Batch Normalization. For a specific configuration in which the encoder design only requires 1Mb, the classification accuracy reaches 87.5% on CIFAR-10. Besides, we also show that this quantized encoder can be used to compress image patch-by-patch while the reconstruction can performed remotely, by a dedicated full-frame decoder. This solution typically enables an end-to-end compression almost without any block artifacts, outperforming patch-based state-of-the-art techniques employing a patch-constant bitrate.