Grid Loss: Detecting Occluded Faces

Abstract

Abstract. Detection of partially occluded objects is a challenging computer vision problem. Standard Convolutional Neural Network (CNN) detectors fail if parts of the detection window are occluded, since not every sub-part of the window is discriminative on its own. To address this issue, we propose a novel loss layer for CNNs, named grid loss, which minimizes the error rate on sub-blocks of a convolution layer independently rather than over the whole feature map. This results in parts being more discriminative on their own, enabling the detector to recover if the detection window is partially occluded. By mapping our loss layer back to a regular fully connected layer, no additional computational cost is incurred at runtime compared to standard CNNs. We demonstrate our method for face detection on several public face detection benchmarks and show that our method outperforms regular CNNs, is suitable for realtime applications and achieves state-of-the-art performance.Keywords: object detection, CNN, face detection (a) Global loss layer.(b) Our grid loss layer.

SSDMNV2: A real time DNN-based face mask detection system using single shot multibox detector and MobileNetV2

“…For uninhibited circumstances, Zhu et al (2017 ) propose a Contextual Multi-Scale Region-based Convolutional Neural Network (CMS-RCNN), which brought a significant impact on the face detection models. To minimize the error on the substitute layers of CNN layers and dealing with the biased obstructions generated in the mask detection models, Opitz et al (2016 ) prepared a grid loss layer. As technology advanced, further CNN-based 3D models started coming up; one was proposed by Li et al (2015 ).…”

“…Secondly, the presence of masks on the face brings a certain kind of noise, which further deteriorates the detection process. These issues have been studied in some existing research papers such as ( Ghiasi & Fowlkes, 2014 ; Opitz, Waltner, Poier, Possegger, & Bischof, 2016 ; Yang, Luo, Loy, & Tang, 2015 ) still, there is an excellent challenge for a vast dataset so that an efficient face mask detection model can be easily developed.…”

A Novel Detection Framework About Conditions of Wearing Face Mask for Helping Control the Spread of COVID-19

“…Recently, a region-based CNN face detector was proposed in [12], which also took the contextual information into account. The work of [13] developed a novel grid loss to solve the occlusion issues in face detection task. For the same purpose, [4] proposed locally linear embedding module to get a similarity-based descriptor.…”

Detecting Lesion Bounding Ellipses with Gaussian Proposal Networks

“…the center coordinates, the major and minor axises and the rotation angle, impact ellipses overlap differently, thus is in spirit similar to our argument. However, in order to maximize the overlap, Opitz et al [19] rasterize both the proposed ellipse and the ground truth ellipse and numerically compute the gradient of each ellipse parameter by counting the change of rasterized overlap in pixels. In comparison, GPN uses KL divergence to analytically optimize the overlap between two Gaussian distributions as a surrogate to optimizing ellipses overlap.…”