improved_CcGAN:用于图像生成的连续条件生成对抗网络

  • c5_228454
    了解作者
  • 4.7MB
    文件大小
  • zip
    文件格式
  • 0
    收藏次数
  • VIP专享
    资源类型
  • 0
    下载次数
  • 2022-04-17 01:20
    上传日期
连续条件生成对抗网络 该存储库提供了我们针对CcGAN的论文中的实验源代码。 如果您使用此代码,请引用 @inproceedings{ ding2021ccgan, title={Cc{GAN}: Continuous Conditional Generative Adversarial Networks for Image Generation}, author={Xin Ding and Yongwei Wang and Zuheng Xu and William J Welch and Z. Jane Wang}, booktitle={International Conference on Learning Representations}, year={2021}, url={https://openrevi
improved_CcGAN-master.zip
内容介绍
# Continuous Conditional Generative Adversarial Networks This repository provides the source codes for the experiments in our papers for CcGANs. <br /> If you use this code, please cite ```text @inproceedings{ ding2021ccgan, title={Cc{GAN}: Continuous Conditional Generative Adversarial Networks for Image Generation}, author={Xin Ding and Yongwei Wang and Zuheng Xu and William J Welch and Z. Jane Wang}, booktitle={International Conference on Learning Representations}, year={2021}, url={https://openreview.net/forum?id=PrzjugOsDeE} } @misc{ding2020continuous, title={Continuous Conditional Generative Adversarial Networks for Image Generation: Novel Losses and Label Input Mechanisms}, author={Xin Ding and Yongwei Wang and Zuheng Xu and William J. Welch and Z. Jane Wang}, year={2020}, eprint={2011.07466}, archivePrefix={arXiv}, primaryClass={cs.CV} } ``` # Hard Vicinal Discriminator Loss (HVDL) and Soft Vicinal Discriminator Loss (SVDL) <p align="center"> </p> An example of the hard vicinity | An example of the soft vicinity :-------------------------:|:-------------------------: ![](images/visualization_HVE.png) | ![](images/visualization_SVE.png) # Naive Label Input (NLI) and Improved Label Input (ILI) Mechanisms NLI for G | NLI for D :-------------------------:|:-------------------------: ![](images/vanilla_label_input_G.png) | ![](images/vanilla_label_input_D.png) CNN for label embedding in ILI | The embedding network in ILI :-------------------------:|:-------------------------: ![](images/pre-trained_CNN_for_label_embedding.png) | ![](images/label_embedding_network.png) ILI for G | ILI for D :-------------------------:|:-------------------------: ![](images/improved_label_input_G.png) | ![](images/improved_label_input_D.png) # 1. Datasets ## The RC-49 Dataset (h5 file) https://1drv.ms/u/s!Arj2pETbYnWQr7MY2Pr5qipSUpZKEQ?e=QTbiq2 <br /> Download 'RC-49_64x64.h5' and put it in './improved_CcGAN/dataset/RC-49' ## The preprocessed UTKFace Dataset (h5 file) https://1drv.ms/u/s!Arj2pETbYnWQr7MW_sGY9tJC4G3eMw?e=ohhRTe <br /> Download 'UTKFace_64x64.h5' and put it in './improved_CcGAN/dataset/UTKFace' ## The Cell-200 dataset (h5 file) https://1drv.ms/u/s!Arj2pETbYnWQr8tDP9Etf16nWddoTQ <br /> Download 'Cell200_64x64.h5' and put it in './improved_CcGAN/dataset/Cell200' ## The Steering Angle dataset (h5 file) For CcGAN, AE, and Regression CNN training: <br /> https://1drv.ms/u/s!Arj2pETbYnWQr7Mdwe6H-IS0YwXh3A?e=U0BiIq <br /> For Clssification CNN training: <br /> https://1drv.ms/u/s!Arj2pETbYnWQr8xEgY3ZHSe2b1CHlQ?e=SE7pv6 <br /> Download 'SteeringAngle_64x64.h5' and 'SteeringAngle_5_scenes_64x64' and put them in './improved_CcGAN/dataset/SteeringAngle' # 2. Sample Usage If a folder has 'improved' in its name, this folder corresponds to a ILI-based CcGAN; otherwise, a NLI-based CcGAN. ## 2.1 Simulation ('./improved_CcGAN/Simulation') First, set the ROOT_PATH and DATA_PATH in the './scripts/run_train.sh' to yours. Then, run 'run_train.sh'. ## 2.2 RC-49 ('./improved_CcGAN/RC-49' and './improved_CcGAN/RC-49-improved'') First, set the ROOT_PATH and DATA_PATH in the './scripts/run_train.sh' to yours. Then, run 'run_train.sh'. ## 2.3 UTKFace ('./improved_CcGAN/UTKFace' and './improved_CcGAN/UTKFace-improved') First, set the ROOT_PATH and DATA_PATH in './scripts/run_train.sh' to yours. Then, run 'run_train.sh'. ## 2.4 Cell-200 ('./improved_CcGAN/Cell200' and './improved_CcGAN/Cell200-improved') First, set the ROOT_PATH and DATA_PATH in './scripts/run_train.sh' to yours. Then, run 'run_train.sh'. ## 2.5 Steering Angle ('./improved_CcGAN/SteeringAngle' and './improved_CcGAN/SteeringAngle-improved') First, set the ROOT_PATH and DATA_PATH in './scripts/run_train.sh' to yours. Then, run 'run_train.sh'. # 3. NIQE The code for computing NIQE is in './improved_CcGAN/NIQE'. A tutorial to compute NIQE for RC-49 will be provided at https://github.com/UBCDingXin/cDRE-based_Subsampling_cGANS/blob/main/RC-49/NIQE/ > Rename the folder containing fake images to fake_images and then compress fake_images with a filename fake_images.zip. Move fake_images.zip to ./RC-49/NIQE/fake_data. Then, run ./RC-49/NIQE/run_test.sh. # 4. Some Results <p align="center"> Line graphs for the RC-49 experiment. </p> <p align="center"> Line graphs for the UTKFace experiment. </p> <p align="center"> Line graphs for the Cell-200 experiment. </p> <p align="center"> Line graphs for the Steering Angle experiment. </p>
评论
    相关推荐
    • Tastebuds:食品订购网站https存储
      食品订购网站的存储库 Tastebuds是一个独特的食品订购网站,您可以在其中自由定制自己的食品。 最好的部分是,您可以根据需要,日期和时间安排下达订单。 味蕾本身就是一个组织,不与其他餐馆捆绑以提供食物。 ...
    • 存储资料
      NULL 博文链接:https://hongyabing.iteye.com/blog/1569369
    • LearningDirectX11:https的源代码存储
      DirectX 12中的纹理和照明 的源代码存储库 控制项 钥匙 行动 W 向前移动相机 A 向左移动相机 S 向后移动相机 D 向右移动相机 Q 向上平移相机 E 向下平移相机 Esc 关闭申请 Alt + Enter 切换全屏模式
    • 存储过程
      NULL 博文链接:https://canglang0.iteye.com/blog/1513344
    • kubemacro-hub:https的宏存储
      KubeMacro集线器 宏存储库。
    • 文档存储
      NULL 博文链接:https://wangjichun.iteye.com/blog/2057313
    • fno-specification:https存储
      存储库 快速开始 编辑dev.html 确保所有本地资产都在resources文件夹中,并且dev.html文件中的链接是相对的(这很重要,因为发布脚本会创建多个嵌套路径) 将快照另存为index.html 运行node publish.js index....
    • commons:此存储库已弃用。 访问 https
      Codenvy 公共类 此存储库已弃用。 访问
    • nur-packages:我的个人NUR存储https
      神经包装 我的个人存储
    • LegacyVEX:https
      这是我用于个人项目的Legacy VEX设备的待添加(TBA)存储库。 如果我的代码可以帮助您完成任务,请随时使用它-如果您使用我的代码的大部分内容,请给我发送代码链接,并在注释部分中将其归功于我,谢谢。 如果您是...