Get all your questions answered about our latest Photosynth Technical Preview.
I really like the gaussian blur used on the backdrop photos when looking at a preview profile page + flipping through the synths.
I'm wondering why you leave the low resolution proxies for the photos while the synth is loading so pixelated.
I could understand if you want people to understand that they ought to expect more to load, however I would like to apply gaussian blurring to all resolutions of photos which are less than 1 pixel of photo per 1 pixel of viewer or more - inside or outside the image pyramid - to the degree that I can't ever perceive the four corners of any pixel when it is being drawn larger than 1 of the screen's pixels.
One thing that I didn't like about the blurring on the photos in the original Photosynth viewers during tile loading was that they were never quite blurred strongly enough to obscure the pixelation.
I wish that the big gradients of a heavily blurred low res image melted into high res with no obvious pop-in or tile edges.
The "big pixels" low-res proxies were a choice by our designer. As you guessed, the goal was to emphasize the fact that "this is not the final view". You're also right that we could have handled this differently, perhaps with a Gaussian blur, but that's an aesthetic issue, and I'm guessing there will be as many opinions as options on that one.
Well the decision is of course yours.
I still think leaving them unblurred is a mistake.
With your history in Seadragon I'm sure that you could give a more in depth talk on why blurring the lower resolutions is smoother visually than I.
Some pros, just for the record:
:: Blurring a low res texture when rendering at a higher resolution:
::: is more natural, in that it looks like something that one's eye might actually see.
::: interpolates values that bridge the gaps between data spatially, artificially increasing the resolution (rendered and perceived) until real higher resolution can be loaded
::: is less high contrast in luminance changes and thus more pleasing to the eye because it is smoother
As background, my preference for blurring low res textures goes back to my teenage years + was cemented in my early twenties so I'm not likely going to ever agree with this particular choice of your designer, even if the photos were all taken perfectly level, however the especially sore points for me are:
:: When playing through any portion of the synth where only the lower resolutions of photos have had time to load in the overwhelming perception for me when viewing these huge pixels onscreen rotated to 'World Up' is, "Error", simply because they are so clearly a grid, but they are misaligned to the screen's grid.
:: Crossfading between two such thumbnails only exacerbates this perception as any two given thumbs are more likely than not to have been been captured at two different 'roll' values.
:: The large square pixels being projected against the higher resolution edges of the depth map's polygon's mattes is very visually displeasing as the grid is shredded.
By contrast, applying enough of a gaussian blur (at render time) to the lower resolution photos completely sidesteps all three of the cons I list above and also blends better with the extracted dominant color being used as the synth's background.
To recap, when blurring the lower resolutions adequately:
:: The user will not be affronted with a lopsided grid onscreen
:: Crossfading between any two photos with different 'roll' is similarly smooth because we are not seeing two misaligned grids
:: The edges of the depth maps will still be quite visible (as they are even using the highest quality photos used during slideshow) but a smooth blur of color trimmed along the crisp edge of a depth map plane is far more visually acceptable.
:: Anti-aliasing becomes less of an issue as well when rendering a blurred thumbnail as filling the screen, as opposed to where the grid of the low res thumbnail and the grid of the screen's pixels are in conflict.
My overall feeling is that as imperfect as the geometry is and all of the '3D artifacts' that occur because of that (as well as the lack of alpha matting along object edges) every step able to be taken to diminish as many visual artifacts as possible ought to be taken - rather than design decisions being made on preference which introduce further artifacts.
I'll stop out of courtesy and recognition that this is your show and you're going to run it as you see fit but I did want to make clear that my objection is not merely preference without reasons - and also give you a strong signal that this is something that will continue to be an objection, regardless of how much more or little it is spoken of.