A multi-scale cascade fully convolutional network face detector

Abstract

Abstract-Face detection is challenging as faces in images could be present at arbitrary locations and in different scales. We propose a three-stage cascade structure based on fully convolutional neural networks (FCNs). It first proposes the approximate locations where the faces may be, then aims to find the accurate location by zooming on to the faces. Each level of the FCN cascade is a multi-scale fully-convolutional network, which generates scores at different locations and in different scales. A score map is generated after each FCN stage. Probable regions of face are selected and fed to the next stage. The number of proposals is decreased after each level, and the areas of regions are decreased to more precisely fit the face. Compared to passing proposals directly between stages, passing probable regions can decrease the number of proposals and reduce the cases where first stage doesn't propose good bounding boxes. We show that by using FCN and score map, the FCN cascade face detector can achieve strong performance on public datasets.

Aid System for Estimating Agricultural Yield Using a Deep Learning Technique: Tomato Case

“…Because, in fact, the density of a tomato is strongly linked to the species [14]. To overcome this problem, studies are underway to set up a mass estimation model based on certain morphological characteristics (extracted from a 2D image of the tomato) for a particular tomato species [19][20][21].…”

A review of small target detection based on deep learning

“…There are many ways to optimize the network structure in addition to feature pyramids for improving the effectiveness of small target detection. Yang et al [21] proposed a three-stage cascaded fully-connected convolutional nerve network. At each level, a score map is generated by a multiscale fully convolutional network that generates scores at different locations and different scales.…”

Extraction and Spatiotemporal Analysis of Impervious Surfaces in Chongqing Based on Enhanced DeepLabv3+

“…However, their drawbacks include slow convergence during training and high requirements for data samples 14 In recent years, deep learning has allowed for signi cant progress to be made in image classi cation, image segmentation, and other elds. Deep learning models can effectively extract both low-level and high-level features from training images, signi cantly enhancing processing speed and accuracy [21][22][23][24] . Traditional convolutional neural networks (CNNs) are mainly used for image classi cation and are suited for xed-size input images, but they struggle with pixel-level semantic segmentation.…”