RootYuan/vit-s-vqgan

More detail in: https://github.com/Qiyuan-Ge/PaintMind

PaintMind

Install

pip install git+https://github.com/Qiyuan-Ge/PaintMind.git

Import

import paintmind as pm

Reconstruction

img = Image.open(img_path).convert('RGB')
img = pm.stage1_transform(is_train=False)(img)
# load pretrained vit-vqgan
model = pm.create_model(arch='vqgan', version='vit-s-vqgan', pretrained=True)
# encode image to latent
z, _, _ = model.encode(img.unsqueeze(0))
# decode latent to image
rec = model.decode(z).squeeze(0)
rec = torch.clamp(rec, -1., 1.)

1.

pm.reconstruction(img_url='https://cdn.pixabay.com/photo/2014/10/22/15/47/squirrel-498139_960_720.jpg')

2.

pm.reconstruction(img_url='https://cdn.pixabay.com/photo/2017/04/09/10/44/sea-shells-2215408_960_720.jpg')

3.

pm.reconstruction(img_url='https://cdn.pixabay.com/photo/2015/06/19/21/24/avenue-815297_960_720.jpg')

4.

pm.reconstruction(img_url='https://cdn.pixabay.com/photo/2017/03/30/18/17/girl-2189247_960_720.jpg')

5.

pm.reconstruction(img_url='https://cdn.pixabay.com/photo/2017/10/28/07/47/woman-2896389_960_720.jpg')