4 ms·
I don't believe 90% of the examples in that video. They perfectly fill the gaps left by the object every single time, including the person's fingerprint, the ca
by imaginenore 12y ago
I don't believe 90% of the examples in that video. They perfectly fill the gaps left by the object every single time, including the person's fingerprint, the carpet pattern, the patterns on the road.
I call BS.
- Joeboy 12y agoThe paper says: """ We compute a mask for the object pixels, and use this mask to inpaint the background using the PatchMatch algorithm [Barnes et al. 2009]. For complex backgrounds, the user may touch up the background image after inpainting. """ So they're not really claiming perfect object removal is part of their thing. What they are actually doing is quite interesting, but their pitch is making some rather extravagant claims IMO.
- imaginenore 12y agoGo ahead and try it. No algorithm will fill in the details like that.
- Joeboy 12y agoI am not disagreeing with you.