search

vinthony / deep-blind-watermark-removal

[AAAI 2021] Split then Refine: Stacked Attention-guided ResUNets for Blind Single Image Visible Watermark Removal
https://arxiv.org/abs/2012.07007
231 stars 55 forks source link

The results before and after running are different #6

Open TomatoBoy90 opened 3 years ago

TomatoBoy90 commented 3 years ago

image

code as follow:

from PIL import Image img=Image.open('XX.jpg').resize((256,256)) image=np.array(img) ih=torch.tensor(np.transpose(image,(2,0,1)))/255.0 ih=ih.unsqueeze(0) imoutput,immask,imwatermark= model(ih.cuda()) imcoarser,imrefine,imwatermark = imoutput[1]immask + im(1-immask),imoutput[0]immask + im(1-immask),imwatermark*immask torchvision.utils.save_image(imcoarser,'imrefine.jpg')

atharvjairath commented 2 years ago

@TomatoBoy90 I guess its because in line imcoarser,imrefine,imwatermark = imoutput[1]immask + im(1-immask),imoutput[0]immask + im(1-immask),imwatermark*immask here you are using im instead of ih.cuda()