carpedm20 / DCGAN-tensorflow

A tensorflow implementation of "Deep Convolutional Generative Adversarial Networks"
http://carpedm20.github.io/faces/
MIT License
7.15k stars 2.63k forks source link

How to create your own dataset? #32

Open DenisSergeevitch opened 8 years ago

DenisSergeevitch commented 8 years ago

I would like to create my own dataset, where I can find information about it? And how is better to auto-crop for 1000 face-photos?

kaihuchen commented 7 years ago

You can find some OpenCV code here for doing the cropping programmingly http://stackoverflow.com/questions/13211745/detect-face-then-autocrop-pictures

I would advice that you also align the center point between the two eyes to the center of the image (the same as in the celebA dataset used in this repository. I have run into problem with training on unaligned images earlier (see my report here for details http://www.terraai.org/avatars/)

shartoo commented 7 years ago

there is a human face dataset named LFW? If you need an open source human face recognize ,the seetaface.

I got a problem,what should my own dataset like?

mohanadkaleia commented 7 years ago

I need to create my own dataset which is not faces, is it just simply provide the path of the dataset during the training phase? When I did that I got the following error: ValueError: Trying to share variable discriminator/d_h3_lin/Matrix, but specified shape (8192, 1) and found shape (25088, 1). Thank you

agulli commented 7 years ago

Anyone was able to create a new dataset? An example would be very useful