DCGAN-pytorch实现带数据集和说明文件_dcgan代码pytorch资源-CSDN下载

共31个文件

pyc：14个

py：8个

png：4个

需积分: 46 113 浏览量 2018-01-07 00:19:46 上传评论 13 收藏 17.35MB ZIP 举报

**DCGAN-pytorch实现详解** DCGAN（Deep Convolutional Generative Adversarial Networks，深度卷积生成对抗网络）是一种基于卷积神经网络（CNN）的生成对抗网络（GAN）架构，它在图像生成任务中表现优秀。PyTorch是一个广泛使用的深度学习框架，以其灵活性和易用性著称，是实现DCGAN的理想工具。本项目提供了完整的DCGAN实现，包括数据集、说明文件以及训练得到的结果图片。 **1. DCGAN的架构** DCGAN的核心在于其生成器（Generator）和判别器（Discriminator）。生成器从随机噪声中生成逼真的图像，而判别器则试图区分真实图像和生成的假图像。在DCGAN中，这两个网络都使用了卷积和反卷积操作，而不是传统的全连接层。 - **生成器**：通常由一系列上采样（upsampling）操作构成，如转置卷积（transpose convolution），将低维随机噪声映射到高维图像空间。 - **判别器**：则包含下采样（downsampling）操作，如卷积，以减少输入图像的尺寸并提取特征，判断输入是真实图像还是生成的假图像。 **2. PyTorch实现** 在PyTorch中，我们可以利用`nn.Module`来构建这两个网络，并使用`torch.optim`中的优化器进行训练。网络的构建主要通过定义卷积和反卷积层，以及激活函数（如ReLU和tanh）。同时，损失函数通常是二元交叉熵损失，用于训练过程中判别器和生成器的优化。 **3. 数据集与预处理** DCGAN通常使用大规模的图像数据集，如MNIST（手写数字）、CIFAR-10或CelebA（名人面部）。在PyTorch实现中，数据集的加载和预处理至关重要，包括图像的归一化、大小调整等步骤，以适应模型的输入要求。 **4. 训练过程** 训练DCGAN涉及交替优化生成器和判别器。固定判别器，优化生成器以生成更真实的图像；接着，固定生成器，优化判别器以更好地识别真实和假图像。这个过程反复进行，直到两个网络达到某种平衡。 **5. 结果展示** 项目提供的结果图片展示了DCGAN生成的图像，可以直观地评估模型的性能。通常，随着训练的进行，生成的图像会越来越接近真实数据的分布。 **6. 说明文件与readme.md** `readme.md`文件通常包含项目介绍、安装步骤、运行代码的说明，以及可能遇到的问题和解决方案。它是理解项目的关键，应仔细阅读以正确使用和复现实验。 DCGAN-pytorch实现项目提供了一个完整的深度学习模型示例，涵盖了生成对抗网络的基础理论和实践应用，对于学习和研究GANs以及PyTorch编程极具价值。通过探索此项目，开发者可以深入理解DCGAN的工作原理，以及如何在PyTorch中实现和训练这样的模型。

资源推荐

资源详情

资源评论

收起资源包目录

pytorch DCGAN.zip （31个子文件）

pytorch DCGAN

logs

celebA_64_64_log.txt 4.1MB

utils

transformed.py 36KB

__pycache__

transformed.cpython-36.pyc 36KB

transformed.cpython-35.pyc 38KB

models

Discriminator.py 1KB

__init__.py 0B

__pycache__

__init__.cpython-35.pyc 122B

Generator.cpython-36.pyc 1KB

Discriminator.cpython-36.pyc 1KB

Generator.cpython-35.pyc 1KB

__init__.cpython-36.pyc 127B

Discriminator.cpython-35.pyc 1KB

Generator.py 1KB

filelist.txt 2.1MB

main.py 11KB

README.md 5KB

data

read_data.py 3KB

filelist.txt 67KB

HumanFaceDataSet.py 2KB

__init__.py 0B

__pycache__

__init__.cpython-35.pyc 120B

HumanFaceDataSet.cpython-36.pyc 2KB

read_data.cpython-35.pyc 2KB

read_data.cpython-36.pyc 2KB

__init__.cpython-36.pyc 125B

HumanFaceDataSet.cpython-35.pyc 2KB

result

result.gif 13.51MB

150.png 762KB

real.png 449KB

50.png 770KB

final.png 753KB

# pytorch-DCGAN-Humanface ## Introduction This is a Pytorch implementation of Deep Convolutional Generative Adversarial Networks (DCGAN) [2] for humanface datasets,witch can genarate some beautiful girls' face. This project is inspired by this blog :https://zhuanlan.zhihu.com/p/26639135 * you can download the dataset from the url below: - HumanFace dataset: https://pan.baidu.com/share/init?surl=dERYUmH - with the password: 09h9 - This dataset contains about 8000 girls‘ face images The main code structure is forked from https://github.com/pytorch/examples/tree/master/dcgan --- ## Implementation details * The input and output image size of the neural network is 64 x 64 , * The learning rate is initialized as 0.0002，and reduces a half every 50 epoch. * The dimension of random noise is set to 1000. * For each iteration, we train the netD first for once and then train the netG for twice. * We trained this network for 300 epoch and got the final result below. --- ## Resutls * Generate using fixed noise (fixed_z_) <table align='center'> <tr align='center'> <td> Training image</td> <td> Result image</td> </tr> <tr> <td><img src = 'result/result.gif'> <td><img src = 'result/final.png'> </tr> </table> * Real image vs Generated images <table align='center'> <tr align='center'> <td> Real image </td> <td> DCGAN after 50 epochs </td> <td> DCGAN after 150 epochs </td> </tr> <tr> <td><img src = 'result/real.png'> <td><img src = 'result/50.png'> <td><img src = 'result/150.png'> </tr> </table> * Learning Time * Humanface DCGAN - about 10 sec per epoch, 300 epoch takes nearly 50 minutes ( If you want to reduce the training time, set the dimension of the noise to 100，then the epoch time is about 7 sec,in my development environment.) * Results * Model: You can find the final trained model in the "checkpoints " folder * Logs: You can get the training log in the "logs " folder --- ## Usage * Data ``` You should download the dataset showed above, and unzip the dataset to the same folder witch contains the main.py file ``` * Run ``` usage: main.py [-h] [--dataset DATASET] [--dataroot DATAROOT] [--train_image_list TRAIN_IMAGE_LIST] [--workers WORKERS] [--batchSize BATCHSIZE] [--imageSize IMAGESIZE] [--nz NZ] [--ngf NGF] [--ndf NDF] [--niter NITER] [--lr LR] [--decay_round DECAY_ROUND] [--beta1 BETA1] [--cuda CUDA] [--ngpu NGPU] [--netG NETG] [--netD NETD] [--outpics OUTPICS] [--outckpts OUTCKPTS] [--outlogs OUTLOGS] [--manualSeed MANUALSEED] optional arguments: -h, --help show this help message and exit --dataset DATASET cifar10 | lsun | imagenet | folder | lfw | fake --dataroot DATAROOT path to dataset --train_image_list TRAIN_IMAGE_LIST pics path lists --workers WORKERS number of data loading workers --batchSize BATCHSIZE input batch size --imageSize IMAGESIZE the height / width of the input image to network --nz NZ size of the latent z vector --ngf NGF number of the filter of generate network --ndf NDF number of the filter of descriminator network --niter NITER number of epochs to train for --lr LR learning rate, default=0.0002 --decay_round DECAY_ROUND learning rate decay 0.5 each decay_round --beta1 BETA1 beta1 for adam. default=0.5 --cuda CUDA enables cuda --ngpu NGPU number of GPUs to use --netG NETG path to netG (to continue training) --netD NETD path to netD (to continue training) --outpics OUTPICS folder to output images --outckpts OUTCKPTS folder to output checkpoints --outlogs OUTLOGS folder to output images --manualSeed MANUALSEED manual seed ``` --- ## Development Environment * Ubuntu 14.04 LTS * NVIDIA GTX 1080 ti * cuda 8.0 * Python 3.5 * pytorch 0.3.0 * torchvision 0.2.0 --- ##Some Reference [1] Goodfellow, Ian, et al. "Generative adversarial nets." Advances in neural information processing systems. 2014. (Full paper: http://papers.nips.cc/paper/5423-generative-adversarial-nets.pdf) [2] Radford, Alec, Luke Metz, and Soumith Chintala. "Unsupervised representation learning with deep convolutional generative adversarial networks." arXiv preprint arXiv:1511.06434 (2015). (Full paper: https://arxiv.org/pdf/1511.06434.pdf) [3] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. "Gradient-based learning applied to document recognition." Proceedings of the IEEE, 86(11):2278-2324, November 1998. [4] Liu, Ziwei, et al. "Deep learning face attributes in the wild." Proceedings of the IEEE International Conference on Computer Vision. 2015.

评论收藏

内容反馈