点击我爱计算机视觉标星,更快获取CVML新技术
图片来自智能体“躲猫猫”
总结了过去一周CV领域的最新开源代码,发现本周出现多份很有价值的高质量、重量级工作,比如致力于使得图卷积网络更深的DeepGCNs、Mask引导的注意力网络大大改进了遮挡行人重识别、格灵深瞳轻量级人脸识别比赛冠军模型VarGFaceNet、比LSTM更优的新RNN模型IndRNN、还有异常强大的字符级文本识别CharNet。
一种web运行的半自动图像标注的灵活框架LOST(Label Objects and Save Time)
LOST: A flexible framework for semi-automatic image annotation
Jonas Jäger, Gereon Reus, Joachim Denzler, Viviane Wolff, Klaus Fricke-Neuderth
https://arxiv.org/abs/1910.07486v1
https://github.com/l3p-cv/lost
对抗表示学习中的全局最优化问题
On the Global Optima of Kernelized Adversarial Representation Learning
Bashir Sadeghi, Runyi Yu, Vishnu Naresh Boddeti
ICCV 2019
https://arxiv.org/abs/1910.07423v1
https://github.com/human-analysis/Kernel-ARL
学习泛化的全尺度表示,用于人员重识别,模型更小,精度更优
Learning Generalisable Omni-Scale Representations for Person Re-Identification
Kaiyang Zhou, Xiatian Zhu, Yongxin Yang, Andrea Cavallaro, Tao Xiang
ICCV 2019
https://arxiv.org/abs/1910.06827v1
https://github.com/KaiyangZhou/deep-person-reid
将ResNet和DenseNet引入到图卷积网络中,可以训练更深(达112层)的GCN,在多个任务中达到了更高的精度。
DeepGCNs: Making GCNs Go as Deep as CNNs
Guohao Li, Matthias Müller, Guocheng Qian, Itzel C. Delgadillo, Abdulellah Abualshour, Ali Thabet, Bernard Ghanem
ICCV 2019
https://arxiv.org/abs/1910.06849v1
https://github.com/lightaime/deep_gcns_torch
https://github.com/lightaime/deep_gcns
训练智能体玩“躲猫猫”游戏
Visual Hide and Seek
Boyuan Chen, Shuran Song, Hod Lipson, Carl Vondrick
https://arxiv.org/abs/1910.07882v1
http://www.cs.columbia.edu/~bchen/visualhideseek/
掩膜引导的注意力网络,用于遮挡严重的行人检测,在多个数据集实现了更高的最好精度。CityPersons提升9.5%,Caltech提升5.0%。
Mask-Guided Attention Network for Occluded Pedestrian Detection
Yanwei Pang, Jin Xie, Muhammad Haris Khan, Rao Muhammad Anwer, Fahad Shahbaz Khan, Ling Shao
ICCV 2019
https://arxiv.org/abs/1910.06160v2
https://github.com/Leotju/MGAN
一种几何启发的卷积操作,有效提升了消失点检测
NeurVPS: Neural Vanishing Point Scanning via Conic Convolution
Yichao Zhou, Haozhi Qi, Jingwei Huang, Yi Ma
https://arxiv.org/abs/1910.06316v1
https://github.com/zhou13/neurvps
单次神经架构搜索,基于自我评估模版网络,在CIFAR和ImageNet数据集达到最先进的性能
One-Shot Neural Architecture Search via Self-Evaluated Template Network
Xuanyi Dong, Yi Yang
ICCV 2019
https://arxiv.org/abs/1910.05733v1
https://github.com/D-X-Y/NAS-Projects
学习鉴别特征,用于非监督域适应
Drop to Adapt: Learning Discriminative Features for Unsupervised Domain Adaptation
Seungmin Lee, Dongwan Kim, Namil Kim, Seong-Gyun Jeong
ICCV 2019
https://arxiv.org/abs/1910.05562v1
https://github.com/postBG/DTA.pytorch
可变组卷积神经网络,可以支持大规模人脸识别,同时减少计算成本和参数。获得格灵深瞳轻量级人脸识别挑战赛冠军!
VarGFaceNet: An Efficient Variable Group Convolutional Neural Network for Lightweight Face Recognition
Mengjia Yan, Mengao Zhao, Zining Xu, Qian Zhang, Guoli Wang, Zhizhong Su
ICCV 2019 Workshop
https://arxiv.org/abs/1910.04985v1
https://github.com/zma-c-137/VarGFaceNet
发明一种称为Hadamard乘积的递归连接,构建了独立递归神经网络(IndRNN),其中同一层中的神经元彼此独立并且跨层连接。
IndRNN可有效替代LSTM,精度更高的同时,速度是其10倍!
Deep Independently Recurrent Neural Network (IndRNN)
Shuai Li, Wanqing Li, Chris Cook, Yanbo Gao, Ce Zhu
https://arxiv.org/abs/1910.06251v1
https://github.com/Sunnydreamrain/IndRNN_pytorch
一种以字符为基本单元的单阶段文本检测识别网络,在三个标准基准上对CharNet结果显示,其结果以最先进的结果大大领先之前的算法,比如ICDAR 2015上从65.33%改进到71.08%,TotalText上从54.0%跃升至69.23%。
Convolutional Character Networks
Linjie Xing, Zhi Tian, Weilin Huang, Matthew R. Scott
ICCV 2019
https://arxiv.org/abs/1910.07954v1
https://github.com/MalongTech/research-charnet
基于语音指令实现的自动驾驶
Conditional Driving from Natural Language Instructions
Junha Roh, Chris Paxton, Andrzej Pronobis, Ali Farhadi, Dieter Fox
CoRL 2019
https://arxiv.org/abs/1910.07615v1
https://sites.google.com/view/language-grounded-driving
医学图像域适应 | 提出了一种新型的无监督域自适应框架,称为协作特征集合自适应(CFEA),改进了眼底图像分割的精度
CFEA: Collaborative Feature Ensembling Adaptation for Domain Adaptation in Unsupervised Optic Disc and Cup Segmentation
Peng Liu, Bin Kong, Zhongyu Li, Shaoting Zhang, Ruogu Fang
MICCAI 2019
https://arxiv.org/abs/1910.07638v1
https://github.com/cswin/AWC
CV细分方向交流群
52CV已经建立多个CV专业交流群,包括:目标跟踪、目标检测、语义分割、姿态估计、人脸识别检测、医学影像处理、超分辨率、神经架构搜索、GAN、强化学习等,扫码添加CV君拉你入群,如已经为CV君其他账号好友请直接私信,
(请务必注明相关方向,比如:目标检测)
喜欢在QQ交流的童鞋,可以加52CV官方QQ群:805388940。
(不会时时在线,如果没能及时通过验证还请见谅)
长按关注我爱计算机视觉