在线时间:8:00-16:00
迪恩网络APP
随时随地掌握行业动态
扫描二维码
关注迪恩网络微信公众号
开源软件名称:TwoStageVAE开源软件地址:https://gitee.com/daib13/TwoStageVAE开源软件介绍:TwoStageVAEThis is the code for the two-stage VAE model proposed in our ICLR 2019 paper "Diagnoising and Enhancing VAE Models" [1]. [1] Dai, B. and Wipf, D. Diagnosing and enhancing VAE models. In International Conference on Learning Representations, 2019. Step 1. Prepare DatasetWe do experiments on MNIST, Fashion-MNIST, Cifar-10 and CelebA dataset. These data are downloaded from the official website and then transformed to npy format using preprocess.py script. Follow the next steps to prepare each dataset. Or you can directly download the data from Google Doc. (If you directly download the data from Google Doc, extract the file to the root folder.) MNISTDownload the data from: http://yann.lecun.com/exdb/mnist/ You will get the files Fashion-MNISTDownload the data from: https://github.com/zalandoresearch/fashion-mnist Again you will get four files To preprocess MNIST and Fashion-MNIST, you also need to install the package pip install python-mnist Cifar-10Download the data (python version) from: https://www.cs.toronto.edu/~kriz/cifar.html Extract the downloaded file in CelebA DatasetDownload the data from: http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html Put the extracted images files (202599 jpg files) in the folder PreprocessTransform the data into python preprocess.py You will obtain some Step 2. Run Two-Stage VAETrain the model by running python demo.py --dataset [DATASET] --network-structure [NETWORK] --exp-name [EXP] --gpu [GPU] The argument
Generated samplesTo reproduce the following results with Resnet architecture, run python demo.py --dataset celeba --epochs 100 --lr-epochs 40 --epochs2 100 --lr-epochs2 40 --network-structure Resnet --num-scale 4 --base-dim 32 --latent-dim 128 --gpu [GPU] --exp-name [EXP]
To reproduce the following results with WAE architecture, run python dome.py --dataset celeba --epochs 70 --lr-epochs 30 --epochs2 70 --lr-epochs2 30 --network-structure Wae --gpu [GPU] --exp-name [EXP]
|
请发表评论