search

Ninebell / GaneratedHandsForReal_TIME

it referenced paper of GANerated Hands for Real-Time 3D Hand Tracking from Monocular RGB.
MIT License
86 stars 17 forks source link

GeoConGAN #11

Closed flyyyyer closed 4 years ago

flyyyyer commented 5 years ago

How can I train the GeoConGAN in my own dataset?

Ninebell commented 5 years ago

Hello sir, I wiil update soon. Anyway, if you want to train the GeoConGAN,

  1. You need to train the Silnet first
  2. Call the sillnet into the GeoConGAN
  3. Set the data path two domain for training.
  4. And start the train~.
flyyyyer commented 4 years ago

My coding is not good, emmm...., so I will be appreciated for you update as soon as possible, thank you very much! I hope you can give a demo to train on the horse2zebra, so I can just change the dataset path to my own dataset.

Ninebell commented 4 years ago

@flyyyyer if you want to train horse2zebra, you need silluet about horse and zebra. Or you mean training cyclegan?

flyyyyer commented 4 years ago

Oh, I just want to train the GeoConGAN on my hand dataset

Ninebell commented 4 years ago

@flyyyyer I had update code about training in here. but readme is not yet.

and you keep the dataset tree like me or you must customize code about calling data.

flyyyyer commented 4 years ago

Thank you for your sharing. In GeoconGan_main.py, the input has the mask image, but I only has real images and synth images. How can I get the mask images? I dont's know how to process the hand RGB images with background to hand mask images.

Ninebell commented 4 years ago

@flyyyyer Hmmm.... I'm sorry about that. My image process code was hard code. and origin real data's mask images had some error. my histories were in here but just read and note only.

flyyyyer commented 4 years ago

So I can't train the GeoconGan if I don't have the mask images.