ECN>> Continuous-Image-Autoencoder>> 返回
项目作者： volotat

项目描述 ：
Deep learning image autoencoder that not depends on image resolution
高级语言： Python
项目地址: git://github.com/volotat/Continuous-Image-Autoencoder.git
创建时间： 2018-01-17T22:34:50Z
项目社区：https://github.com/volotat/Continuous-Image-Autoencoder
开源协议：
下载


Continuous-Image-Autoencoder

Deep learning image autoencoder that not depends on image resolution

How to use:

You should provide path to dataset file and number of examples in it to train new model, like so:

  1. CIA.py -f cars.png -e 20

In dataset image there should be 10 pictures in row and as many rows as you want. By default each picture individually expect to have 200x200 resolution, but you can change it through “size” parameter in script if it’s necessary.
If you want to generate new images with pretrained model just add “—no-train” parameter.
Be patient while image generating, its might take quite a time.

Model structure:

model_structure

Results example:

  • Cars: 20 examples, 3D Latent space
    exmpls_cars
  • Faces: 30 examples, 5D Latent space
    exmpls_faces

You will find more examples in folders in “out.png” files.