Network quantization allows inference to be conducted using low-precision arithmetic for improved inference efficiency of deep neural networks on edge devices. However, designing aggressively low-bit (e.g., 2-bit) quantization schemes on complex tasks, such as object detection, still remains challenging in terms of severe performance degradation and unverifiable efficiency on common hardware. In this paper, we propose an Accurate Quantized object Detection solution, termed AQD, to fully get rid of floating-point computation. To this end, we target using fixed-point operations in all kinds of layers, including the convolutional layers, normalization layers, and skip connections, allowing the inference to be executed using integer-only arithmetic. To demonstrate the improved latency-vs-accuracy trade-off, we apply the proposed methods on RetinaNet and FCOS. In particular, experimental results on MS-COCO dataset show that our AQD achieves comparable or even better performance compared with the full-precision counterpart under extremely low-bit schemes, which is of great practical value. Source code and models are available at: https://github.com/ziplab/QTool

翻译：网络量化使得推理可以使用低精度算术进行，从而提升深度神经网络在边缘设备上的推理效率。然而，针对目标检测等复杂任务设计激进的低比特（如2比特）量化方案，仍面临性能严重下降及在常见硬件上效率无法验证的挑战。本文提出一种名为AQD的精确量化目标检测解决方案，以完全摆脱浮点计算。为此，我们目标是所有类型的层（包括卷积层、归一化层和跳跃连接）均采用定点运算，使得推理仅通过整数算术执行。为展示延迟与精度权衡的改善，我们将所提方法应用于RetinaNet和FCOS。特别地，在MS-COCO数据集上的实验结果表明，在极低比特方案下，我们的AQD相比全精度对应方法达到了相当甚至更优的性能，这具有重要的实用价值。源代码和模型可在 https://github.com/ziplab/QTool 获取。