lzw-all-in / Anime_person_translation
人脸和动漫脸的互转
Stars: ✭ 35
Programming Languages
Projects that are alternatives of or similar to Anime person translation
Cyclegan Qp
Official PyTorch implementation of "Artist Style Transfer Via Quadratic Potential"
Stars: ✭ 59 (+68.57%)
Mutual labels: gan, cyclegan
Cyclegan Vc2
Voice Conversion by CycleGAN (语音克隆/语音转换): CycleGAN-VC2
Stars: ✭ 158 (+351.43%)
Mutual labels: gan, cyclegan
Cyclegan tensorlayer
Re-implement CycleGAN in Tensorlayer
Stars: ✭ 86 (+145.71%)
Mutual labels: gan, cyclegan
Pytorch Cyclegan And Pix2pix
Image-to-Image Translation in PyTorch
Stars: ✭ 16,477 (+46977.14%)
Mutual labels: gan, cyclegan
Cyclegan Tensorflow
An implementation of CycleGan using TensorFlow
Stars: ✭ 1,096 (+3031.43%)
Mutual labels: gan, cyclegan
Cyclegan Vc3
Voice Conversion by CycleGAN (语音克隆/语音转换)：CycleGAN-VC3
Stars: ✭ 52 (+48.57%)
Mutual labels: gan, cyclegan
Paddlegan
PaddlePaddle GAN library, including lots of interesting applications like First-Order motion transfer, wav2lip, picture repair, image editing, photo2cartoon, image style transfer, and so on.
Stars: ✭ 4,987 (+14148.57%)
Mutual labels: gan, cyclegan
Cyclegan Music Style Transfer
Symbolic Music Genre Transfer with CycleGAN
Stars: ✭ 201 (+474.29%)
Mutual labels: gan, cyclegan
Cyclegan
Software that can generate photos from paintings, turn horses into zebras, perform style transfer, and more.
Stars: ✭ 10,933 (+31137.14%)
Mutual labels: gan, cyclegan
Pytorch-Image-Translation-GANs
Pytorch implementations of most popular image-translation GANs, including Pixel2Pixel, CycleGAN and StarGAN.
Stars: ✭ 106 (+202.86%)
Mutual labels: gan, cyclegan
Gannotation
GANnotation (PyTorch): Landmark-guided face to face synthesis using GANs (And a triple consistency loss!)
Stars: ✭ 167 (+377.14%)
Mutual labels: gan, cyclegan
automatic-manga-colorization
Use keras.js and cyclegan-keras to colorize manga automatically. All computation in browser. Demo is online:
Stars: ✭ 20 (-42.86%)
Mutual labels: gan, cyclegan
Pycadl
Python package with source code from the course "Creative Applications of Deep Learning w/ TensorFlow"
Stars: ✭ 356 (+917.14%)
Mutual labels: gan, cyclegan
Nag
[CVPR 2018] Tensorflow implementation of NAG : Network for Adversary Generation
Stars: ✭ 24 (-31.43%)
Mutual labels: gan
Advanced Deep Learning With Keras
Advanced Deep Learning with Keras, published by Packt
Stars: ✭ 917 (+2520%)
Mutual labels: gan
Dcgan Pytorch
PyTorch Implementation of DCGAN trained on the CelebA dataset.
Stars: ✭ 32 (-8.57%)
Mutual labels: gan
人脸与动漫脸互转
基本说明
- 代码中主要尝试了2种框架 cycleGAN 和 DTN 但是DTN的效果太差甚至跑不起来
- 尝试了各种优化方式比如：WGAN 、WGAN-GP 、SN 最后得出结论虽然SN好与前两者，但是存在很严重的mode collapse的问题，就是无论输入什么样的人脸输出的动漫脸都是同一张。虽然WGAN能跑起来也朝着自己想要的方向发展但是WGAN真的跑的太慢了>_<，40轮和80轮变化微小还跑了整整一天 ==\，不光是每个epoch的耗时更长而且模型收敛的也很慢啊～～～～
- 最后在cycleGAN的框架之上采用了self-attention GAN 才最终解决问题，也就是我的最终版本
- 数据集：人脸数据是 CelebA ，动漫脸数据是 知乎上有人已经爬取下来的 密码：g5qa 。不过自己只使用了前面50000张图片
- 所有图片在实际运行时都转化为脸64x64的尺寸，最终的模型训练了200轮。自己借了同学实验室4张K20的显卡来跑(表示没有显卡完全搞不动深度学习//(ㄒoㄒ)//)跑了3天多吧(自己记得没错的话)。
- 整个cycleGAN框架的代码很大一部分是来源于论文源码，自己在其基础上进行修改的，虽然很想说是自己写的，不过还是应该感谢一下原论文的作者！b（￣▽￣）d
文件说明
- 每个文件的功能基本上和文件名相对应。
model_ckpt存放的是预训练的模型，如果要使用模型直接运行test.py就可以使用了(前提是argments.py里面图片路径是正确的)。save_image是我之前跑模型时存下来看效果的图片结果后面的数字代表训练集的大小前面的英文代表GAN使用的优化方法。test_image是最后使用预训练模型得到的测试结果。argments.py里面存放了模型的各种配置参数，当然也有些参数没有放入，比如训练时所使用的GPU的id - argument.py里面存放了大部分的参数设置，最后的结果就是直接使用默认配置跑出来的，如果需要从头跑的话只需要修改图片路径(
domain_A_path,domain_B_path)和cyclegan_model.py里面根据自己的gpu个数修改self.gpu_ids哦！
实验环境说明
- python 3.5
- pytorch 0.4
- 4块K20显卡
效果展示
真实人脸
人脸转化的动漫脸
真实动漫脸
动漫脸转化的人脸
自我评价
感觉虽然动漫脸和人脸的转换的结果还是很不错，但是人物和动漫人物却没有很相似的成分(个人认为)，并且不知道是不是因为动漫的数据集女性偏多的原因所以所有男性人脸都转化到了女性的动漫脸上了╮(￣▽￣)╭
需要看更多样片的请进入test_image文件另外欢迎有任何问题的可以给我提issue哦！就酱！∩__∩y ！
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].