Cliping issue

[Vaalizaadeh]

Hey, man! Thanks for your implementation. I guess there is something that has to be noticed. Don't you think you have to clip the x in your bilinear function, too?

[kevinzakka]

@Vaalizaadeh I don't think it matters because I end up clipping the 4 corner coordinates before sampling. So suppose a certain coordinate lies outside the image range.

• If I clip x before grabbing the 4 corner points, I'll have that x0 = max_x and x1 = max_x + 1 after which I'll clip again and obtain x0 = x1 = max_x.
• If I don't clip before (like in the current implementation), I'll have that x0 = some_value_greater_than_max_x and x1 = some_value_greater_than_max_x + 1 after which I'll clip again and obtain x0 = x1 = max_x.

So no difference.

[Vaalizaadeh]

@kevinzakka I guess if you don't clip, the intensity of boundary locations will have wrong calculations. Take a look at the weights of the interpolation line. suppose one x which is not clipped has a value equal to $255.5$ and the range is $255$. it may lead to wrong answers if the other coordinate lies in the plane.