Pruned Lightweight Encoders for Computer Vision

TL;DR

This paper proposes using pruned lightweight encoders like ASTC and JPEG XS for low-latency computer vision, demonstrating that retraining neural networks with compressed data can recover accuracy losses caused by compression artifacts.

Contribution

It introduces a retraining approach to mitigate accuracy degradation from lightweight, pruned encoders without altering network architecture or coding formats.

Findings

ASTC encoder is 2.3x faster than JPEG.

Retraining reduces accuracy loss to below 5 percentage points.

Disabling significance flag coding in JPEG XS saves 22-23% encoding time.

Abstract

Latency-critical computer vision systems, such as autonomous driving or drone control, require fast image or video compression when offloading neural network inference to a remote computer. To ensure low latency on a near-sensor edge device, we propose the use of lightweight encoders with constant bitrate and pruned encoding configurations, namely, ASTC and JPEG XS. Pruning introduces significant distortion which we show can be recovered by retraining the neural network with compressed data after decompression. Such an approach does not modify the network architecture or require coding format modifications. By retraining with compressed datasets, we reduced the classification accuracy and segmentation mean intersection over union (mIoU) degradation due to ASTC compression to 4.9-5.0 percentage points (pp) and 4.4-4.0 pp, respectively. With the same method, the mIoU lost due to JPEG XS compression at the main profile was restored to 2.7-2.3 pp. In terms of encoding speed, our ASTC encoder implementation is 2.3x faster than JPEG. Even though the JPEG XS reference encoder requires optimizations to reach low latency, we showed that disabling significance flag coding saves 22-23% of encoding time at the cost of 0.4-0.3 mIoU after retraining.