one-shot learning && GAN的资料汇集

0. 写作目的

好记性不如烂笔头。

1. one shot learning 

1. VALSE2018 有关One-shot-learning 的报告:https://www.jiqizhixin.com/articles/2018-06-20-13

有关one-shot-learning的博客:1) http://chuansong.me/n/2131635951225

       2) https://blog.csdn.net/qq_16234613/article/details/79902085

2. Zero-shot-learning 

虽然写的Few-shot learning,但是做实验时,作者做的是Zero-shot learning的实验。

ECCV2018: Semantic Feature Augmentation in Few-shot learning

 论文中给出one-shot data augmentation的六种方法:

i) Learning one-shot models by utilizing the man-ifold information of large amount of unlabelled data in a semi-supervised or transductive setting 通过流信息使用未标注的图像进行训练

ii) Adaptively learning the one-shot classiers from off shelf trained models, 依据现有的模型进行学习(特指分类任务)

iii) Borrowing examples from relevant categories or semantic vocabularies to augment the training set, 从相近的类别或者语义词汇来增广

iv) Synthesizing new labelled training data by rendering virtual examples or composing synthesized representations or distorting existing training examples,合成新图像使用一般的图像增广方法

v) Generating new examples using Generative Adversarial Networks (GANs), 用GAN进行增广

vi) Attribute-guided augmentation (AGA) to synthesize samples at desired values or strength,利用属性集,词向量来合成,如本文中,在语义空间进行增广

2. GAN用于数据增广

2.1 GAN的学习资料

GAN的调研: https://zhuanlan.zhihu.com/p/32103958

GAN的各种paper汇集(包括Generating High-Quality Images, Object Detection,Image Translation):https://zhuanlan.zhihu.com/p/42606381

独家 | GAN大盘点,聊聊这些年的生成对抗网络 : LSGAN, WGAN, CGAN, infoGAN, EBGAN, BEGAN, VAE

NVIDIA 生成高清的图像

GAN-ZOO: https://github.com/hindupuravinash/the-gan-zoo

 

 

你可能感兴趣的:(One-shot,Learning,Zero-shot,learning,GAN,GAN学习,One-shot,learning,for,classifi)