Summer Tree is Cyan

Thinking will not overcome fear but action will.

距离和方差对标签估计准确率的影响分析

方差实验组

通用说明 model0-model10,训练程度加深。 横坐标为未标注样本。 distance 指的是各个样本的与带标注样本的最小欧几里得距离。 variance 指的是各个样本与其距离最近的两个带标注样本的距离方差。 FN_acc(n<U)表示前n个未标注样本标签估计的准确率。其中U=len(u_data) acc_list 是bool数值,长度为U, 如果...

ZstGAN | An Adversarial Approach forUnsupervised Zero-Shot Image-to-Image Translation

Abstract In this work In this workwe, we propose a framework calledZstGAN: By introducing an adversarial training scheme,ZstGAN learns to model each domain with domain-specificfeature distribution ...

Weakly Supervised Classification | Towards Accurate Machine Learning with Low Labeling Costs

Masashi Sugiyama | 弱监督机器学习研究进展

reference link : HPCL-智能计算 · NUDT Title: Weakly Supervised Classification: Towards Accurate Machine Learning with Low Labeling Costs 报告人:Prof. Masashi Sugiyama, The University of Tokyo 报告摘...

Decoupled Novel Object Captioner

(image-text) Abstract In this paper, we introduce the zero-shot novel object caption-ing task where the machine generates descriptions without extratraining sentences about the novel object. To ta...

Generative Adversarial Text to Image Synthesis

Abstract In this work, we develop a novel deeparchitecture and GAN formulation to effectivelybridge these advances in text and image model-ing, translating visual concepts from charactersto pixels....

Dynamic Label Graph Matching for Unsupervised Video Re-Identification

Abstract This pa-per focuses on cross-camera label estimation, which can be subsequently used in feature learning to learn robust re-IDmodels. Specifically, we propose to construct a graph forsamp...

Stepwise Metric Promotion for Unsupervised Video Person Re-identification

Abstract two assumptions two assumptions 1) different video track-lets typically contain different persons。 2)within each tracklet, the frames are mostly of the same per-son. \ Our method is built...

Image-Image Domain Adaptation with Preserved Self-Similarity and Domain-Dissimilarity for Person Re-identification

main methords In our at-tempt, we present a “learning via translation” framework.In the baseline, we translate the labeled images from sourceto target domain in an unsupervised manner. we propose ...

Re-ID 2019 Review

Academic Report

Train/Test on the same domain Part-level features are effective. pose and 3D information are beneficial **Smothness in feature space is beneficial ** Unsupervised domain adaptation Evolutio...

为什么-关于因果关系的新科学 | 01 因果关系之梯

因果推断系列丛书

上帝问的是“什么”,他们回答的却是“为什么”。上帝询问事实,他们回答理由。 而且,两人都深信,列举原因可以以某种方式美化他们的行为。他们是从哪里得到这样的想法的? 人类祖先想象不存在之物的能力是一切的关键,正是这种能力让他们得以交流得更加顺畅。在获得这种能力之前,他们只相信自己的直系亲属或者本部落的人。而此后,信任就因共同的幻想 (例如信仰无形但可想象的神,信仰来世,或者信仰领袖的神性)和...