Deep Learning Face Representation from Predicting 10,000 Classes论文笔记(2015.03.24) 一.基本思路 作者利用卷积神经网络(Convolutional Neural Network,CNN)对大量样本进行训练,提取Deep hidden identity feature(DeepID)特征,然后利用这些特征进行人脸验证(Face Verification).在LFW(Labeled Faces in the Wild)库上
Background 1) "Patch-level image representation"的优势 "Patch-level image representation is very important for object classification and detection, since it is robust to spatial transformation, scale variation, and cluttered background" &
Deep Reinforcement Learning for Visual Object Tracking in Videos 论文笔记 arXiv 摘要:本文提出了一种 DRL 算法进行单目标跟踪,算是单目标跟踪中比较早的应用强化学习算法的一个工作. 在基于深度学习的方法中,想学习一个较好的 robust spatial and temporal representation for continuous video data 是非常困难的. 尽管最近的 CNN based tracke