行人重识别:
1、Bag of Tricks and A Strong Baseline for Deep Person Re-identification
地址:http://103.95.217.77/openaccess.thecvf.com/content_CVPRW_2019/papers/TRMTMCT/Luo_Bag_of_Tricks_and_a_Strong_Baseline_for_Deep_Person_CVPRW_2019_paper.pdf
描述:显而易见,在行人重识别中的一些小技巧
2、Adaptive deep metric embeddings for person re-identification under occlusions(Neurocomputing)
地址:https://www.researchgate.net/publication/331323968_Adaptive_Deep_Metric_Embeddings_for_Person_Re-Identification_under_Occlusions
描述:利用LSTM提取局部信息,并结合全局信息、提出triplet loss的改进版本ANN loss。
3、Diversity-Achieving Slow-DropBlock Network for Person Re-Identification
地址:http://103.95.217.75/xxx.itp.ac.cn/pdf/2002.04414.pdf
描述:一个是slow-dropblock,另一个就是注意力机制和支路的使用、GAP提取全局信息,GMP提取局部信息。
图神经网络
1、A Comprehensive Survey on Graph Neural Networks
地址:https://arxiv.org/abs/1901.00596?context=cs
描述:图神经网络综述。
细粒度图像分类
1、The Devil is in the Channels: Mutual-Channel Loss for Fine-Grained Image Classification
描述:提出了一种多通道损失
GAP,CCMP,和CWA分别是全局平均池化(global average pooling),跨通道最大池化(cross-channel max pooling)和通道维度的注意力机制(channel-wise attention)的简写
数据增强
1、GradAug: A New Regularization Method for Deep Neural Networks
地址:http://xxx.itp.ac.cn/pdf/2006.07989.pdf
小目标检测
1、HRDNet: High-resolution Detection Network for Small Objects
地址:http://xxx.itp.ac.cn/pdf/2006.07607.pdf
自然语言处理
1、On Layer Normalization in the Transformer Architecture
地址:https://arxiv.org/pdf/2002.04745v1.pdf
行人检测
1、YOLOPEDS: EFFICIENT REAL-TIME SINGLE-SHOT PEDESTRIAN DETECTION FOR SMART CAMERA APPLICATIONS