search

YuYin1 / DA-GAN

46 stars 14 forks source link

trained on lfw #4

Open DamDam123 opened 4 years ago

DamDam123 commented 4 years ago

Hello, I want to ask how do you train and test the LFW dataset? I found that some people in the lfw have only one image, and others with multiple images how to set the ground truth and the side input image for training and testing? Hope to reply, thanks

DamDam123 commented 4 years ago

Hello, I'm disturbing you again. I tried to preprocess the lfw data set, but the cropping effect is not good. Can you send me the processed lfw data set? If you can, I would appreciate it！

csyxwei commented 4 years ago

In previous work, LFW is only used to test the model, and the model is trained on Multi-PIE. So I guess DA-GAN has the same training and testing process.

LightCNN has provided the aliged LFW, but they are grayscale images. If you want to align image by yourself, you can use the OpenFace toolbox to extract the landmarks, and then align image according to the method provided by TP-GAN. Since there may be multiple faces in one image, and OpenFace will extract the key points of all faces. You may need to choose the correct landmarks to align the image. Generally, the face in the middle of the image is the one corresponds to the label.

DamDam123 commented 4 years ago

In previous work, LFW is only used to test the model, and the model is trained on Multi-PIE. So I guess DA-GAN has the same training and testing process.

LightCNN has provided the aliged LFW, but they are grayscale images. If you want to align image by yourself, you can use the OpenFace toolbox to extract the landmarks, and then align image according to the method provided by TP-GAN. Since there may be multiple faces in one image, and OpenFace will extract the key points of all faces. You may need to choose the correct landmarks to align the image. Generally, the face in the middle of the image is the one corresponds to the label.

Thank you very much for your reply, I saw your FFWM, but I did not find the paper "Learning Flow-based Feature Warping for Face Frontalization with Illumination Inconsistent Supervision", I am interested in this, I hope you will open source the code soon.Thanks!