Why the input of network is torch.cat([masked_img, (1-mask)]).

NiFangBaAGe / CoordFill

[AAAI 2023] CoordFill: Efficient High-Resolution Image Inpainting via Parameterized Coordinate Querying

BSD 3-Clause "New" or "Revised" License

84 stars 7 forks source link

Why the input of network is torch.cat([masked_img, (1-mask)]). #12

Open LoveU3tHousand2 opened 9 months ago

LoveU3tHousand2 commented 9 months ago

The input of in-painting network always was torch.cat([masked_img, mask]) in early works, why it is reversed in coordfill ? I had spend a lot time to find this 'bug'... Is there any reason?