On 9/11/14 3:22 PM, cowwoc wrote:
> Here is my use-case: I've got this 4x3 section in my UI for video.
> Now, I want the highest resolution video you can give me with this
> aspect ratio. If you return some low-res video I will up-scale it to
> fit. Meaning, 800x1200 or 400x600 are fine but 788x1200 is not because
> it will look bad to have a column of 12 black pixels alongside the
> video (I'm aiming for no black bars or cropping). If I really have to
> (the webcam doesn't support the desired resolution) I'll crop the
> video, but I don't like you making me crop the video when I don't have
> to (rounding error caused by the epsilon value).
>
> I hope this clears it up.
Thanks, that helps! - First, I note that your example doesn't seem to be
a problem with Harald's value, so there's that.
Being precise surely helps, though realistically, what camera has two
resolutions within the margin of error of a two- (or for Harald's value
three)-digit aspect-ratio?
But let me assume there is one...
I may be wrong on this, but if we *did* decide on just two decimals of
precision on inputs, then rather than round everything on input - which
might cause error-accumulation - I'd assumed we'd instead round
non-destructively inside some comparison function used by min, max and
exact. That way it seems possible to me that ideal's distance algorithm
could still work reliably enough to give 800x1200 a smaller distance
than 788x1200, provided an { ideal: 4/6 }. Epsilon is perhaps not the
best word for this, and I may not be making any sense, and it sounds a
bit complicated.
.: Jan-Ivar :.
Received on Thursday, 11 September 2014 20:39:00 UTC