论文笔记:Precomputed Real-Time Texture Synthesis with Markovian Generative Adversarial Networks
2016-08-08 16:19
656 查看
1.什么是deconvolution
反向映射,用来可视化深度特征,也可以用来重建图片。
2.合成图片两种方式:
1)生成全图的模型,效果不错但只对小图work,保真度fidelty低,一般用auto encoder
2)马尔科夫模型,同时生成texture。可以捕获局部patch的统计信息。本文是第二种
3.主要通过strided convolutional network取代pooling来加速inversion of the network
motivation:真实数据往往不是正态分布的,而是一个非线性manifold,用高斯去mapping效果不好,我们学习跟那个manifold的内容相关的patch的mapping。
4.network
loss
训练过程不改变vgg19,只通过优化D,G来最大化G的质量。
文章最后对比了不用vgg做finetune的效果很差。
反向映射,用来可视化深度特征,也可以用来重建图片。
2.合成图片两种方式:
1)生成全图的模型,效果不错但只对小图work,保真度fidelty低,一般用auto encoder
2)马尔科夫模型,同时生成texture。可以捕获局部patch的统计信息。本文是第二种
3.主要通过strided convolutional network取代pooling来加速inversion of the network
motivation:真实数据往往不是正态分布的,而是一个非线性manifold,用高斯去mapping效果不好,我们学习跟那个manifold的内容相关的patch的mapping。
4.network
loss
训练过程不改变vgg19,只通过优化D,G来最大化G的质量。
文章最后对比了不用vgg做finetune的效果很差。
相关文章推荐
- StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks 论文笔记
- Real-Time Video Super-Resolution with Spatio-Temporal Networks and Motion Compensation 论文笔记
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks----论文笔记
- 【论文笔记】Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
- 论文笔记| Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
- 论文笔记之:Semi-Supervised Learning with Generative Adversarial Networks
- 论文笔记之:UNSUPERVISED REPRESENTATION LEARNING WITH DEEP CONVOLUTIONAL GENERATIVE ADVERSARIAL NETWORKS
- 论文笔记:Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks
- 论文笔记:Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks
- 论文笔记:Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks
- 论文笔记:unsupervised representation learning with deep convolutional generative adversarial networks
- 论文阅读笔记:Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
- 论文笔记:Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks
- 论文笔记-深度估计(7)-CNN-SLAM Real-time dense monocular SLAM with learned depth prediction
- 论文笔记:Spectral Normalization for Generative Adversarial Networks [ICLR2018 oral]
- 论文翻译:Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
- Face Aging with Conditional Generative Adversarial Network 论文笔记
- 生成对抗网络学习笔记3----论文unsupervised representation learning with deep convolutional generative adversarial
- 论文笔记:PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
- 论文笔记之:Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks