HarDNet: A Low Memory Traffic Network

By Ping Chao, Chao-Yang Kao, Yunxing Ruan, Chien-Hsiang Huang, Youn-Long Steve Lin. IEEE International Conference on Computer Vision, 2019

307 citations recorded by the source corpus.

State-of-the-art neural network architectures such as ResNet, MobileNet, and DenseNet have achieved outstanding accuracy over low MACs and small model size counterparts. However, these metrics might not be accurate for predicting the inference time. We suggest that memory traffic for accessing intermediate feature maps can be a factor dominating the inference latency, especially in such tasks as real-time object detection and semantic segmentation of high-resolution video. We propose a Harmonic Densely Connected Network to achieve high efficiency in terms of both low MACs and memory traffic. The new network achieves 35%, 36%, 30%, 32%, and 45% inference time reduction compared with FC-DenseNet-103, DenseNet-264, ResNet-50, ResNet-152, and SSD-VGG, respectively. We use tools including Nvidia profiler and ARM Scale-Sim to measure the memory traffic and verify that the inference latency is indeed proportional to the memory traffic consumption and the proposed network consumes low memory traffic. We conclude that one should take memory traffic into consideration when designing neural network architectures for high-resolution applications at the edge.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Electrical and Electronic Engineering

View Paper: https://www.semanticscholar.org/paper/0c55b3a00349e8e8c38ae3f7f82405342df03b2a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Electrical and Electronic Engineering

Citations in the corpus

Polyp-PVT: Polyp Segmentation with Pyramid Vision Transformers (2021, methodology): “Specifically, they used Hardnet [34] to replace the original backbone network Res2Net50 backbone network and removed the attention mechanism to achieve faster and more accurate polyp segmentation.”
Scaled-YOLOv4: Scaling Cross Stage Partial Network (2020, background): “Minimize Convolutional Input/Output (CIO): CIO [4] is an indicator that can measure the status of DRAM IO.”
YOLOv4: Optimal Speed and Accuracy of Object Detection (2020, background): “HarDNet: A Low Memory Traffic Network [4] RFBNet HarDNet68 512 41.”
Towards the Systematic Reporting of the Energy and Carbon Footprints of Machine Learning (2020, methodology): “…models available through PyTorch Hub (see https://pytorch.org/hub )—namely (Krizhevsky et al., 2012; Huang et al., 2017; Szegedy et al., 2015; Chao et al., 2019; Sandler et al., 2018; Zhang et al., 2018; Iandola et al., 2016; Simonyan and Zisserman, 2014; Zagoruyko and Komodakis, 2016).”
FairMOT: On the Fairness of Detection and Re-identification in Multiple Object Tracking (2020, methodology): “, 2019a), HarDNet (Chao et al., 2019) and RegNet (Radosavovic et al.”
PP-YOLO: An Effective and Efficient Implementation of Object Detector (2020, unknown): “5% RFBNet[3] HarDNet85 512 37.”
CSPNet: A New Backbone that can Enhance Learning Capability of CNN (2019, methodology): “[2] proposed a low memory traffic CNN called Harmonic DenseNet (HarDNet) and a metric Convolutional Input/Output (CIO) which is an approximation of DRAM traffic proportional to the real DRAM traffic measurement.”
SegDiff: Image Segmentation with Diffusion Probabilistic Models (2021, methodology): “The leading methods in the ﬁeld employ encoder-decoder networks of varied architectures [4, 31, 38, 50, 52, 53].”
A review on modern defect detection models using DCNNs – Deep convolutional neural networks (2021, unknown): “…MobileNetV2 [71] 25.7% 3 YOLOv3-608 [31] 33.0% 20 SSD512-HarDNet85 [72] 35.1% 39.0 Mask R-CNN X-152-32x8d [73] 40.3% 3 YOLOv4-608 [63] 43.5% 62.0 CenterNet HarDNet-85 [72] 43.6% 45.0 SpineNet-49 [74] 45.3% 29.1 NAS-FPN AmoebaNet [71] 48.3% 3.6 EfﬁcientDet-D7x(single-scale) [69] 55.1% 6.5”