HarDNeXt：基於區段感受視野和連結重要性建構的卷積神經網路

具有並列連接的最新卷積神經網絡，例如 DenseNet，FractalNet 和 HarDNet 藉由大量的特徵融合運算在眾多計算機視覺任務（包括圖像分類和目標檢測）取得了傲人的成績。但是，相對密集的連接模式限制了這些有效模型的推理速度。在本文中，我們提出了一個維持多尺度特徵融合特色且但具有更少的記憶體搬運量的新計算模塊，稱為 HarDX 塊和一個用於設計特定於輸入大小的卷積神經網路結構名為 Stage Receptive Field 的新穎概念。基於這兩個創新的概念，我們提出了一種新的卷積神經網路結構，名為 HarDNeXt，速度分別比 DenseNets 和 HarDNets 快了 200％和 30％並且在進行推論所需的功耗上分別少了 60％和 15％。對於結腸鏡檢查圖像息肉分割的應用任務，HarDNeXt 與最先進的神經網路模型相比達到相同水平的精度並將運行速度提高了 25％。

關鍵字

深度學習；卷機神經網路；節能；感受視野

並列摘要

Stateoftheart Convolution Neural Networks with extensive feature aggregation via shortcut connections such as DenseNet, FractalNet and HarDNet have achieved remarkable results with extensive feature aggregations in numerous computer vision tasks, including image classification and object detection . However, the relatively dense connection pattern has impaired the inference speed. We propose a new computation block, called HarDX block, for maintaining multireceptive field feature fusion with low DRAM traffic and a novel concept, called Stage Receptive Field, for designing inputsizespecific network architectures. Accordingly, we propose a new CNN architecture named HarDNeXt, which is 200% and 30% faster than DenseNets and HarDNets, respectively, while consuming 60% and 15% less energy, respectively. For an application task of polyp segmentation of colonoscopy images, HarDNeXt achieves the same level of accuracy and runs 25% faster compared with a stateoftheart network. Code, data, and experiment setup are opensourced in GitHub.