options.py in glasses dataset

chenhsuanlin / spatial-transformer-GAN

ST-GAN: Spatial Transformer Generative Adversarial Networks for Image Compositing :eyeglasses: (CVPR 2018)

MIT License

334 stars 72 forks source link

options.py in glasses dataset #30

Closed ArvindKumarS closed 5 years ago

ArvindKumarS commented 5 years ago

The dataH and dataW are set to 218 nd 178 opt.dataH,opt.dataW = 218,178 opt.centerY,opt.centerX = 124,89 But the foreground glasses are 144 x 144. What exactly is the 218 and 178 ?

chenhsuanlin commented 5 years ago

The size of the raw CelebA images (aligned version) is 218x178, and we crop 144x144 images around the faces centered at (124,89). The glasses are thus composited with the cropped 144x144 CelebA faces. Hope this clarifies.