/vr/ - Retro Games

>>2330197
>If you want to get really technical, the PS1 wasn't even a fully 3D machine (textures are not 3D perspective correct).
Meaningless technobabble. Flat-shaded and wireframe games are equally 3D, because 3D information is projected onto the screen. Affine texture mapping, while ugly, is not some smoking gun that changes the whole definition. Anyway, it's far from a black mark against the PSX and Saturn, since it could be (and was) fought with higher tesselation. Plenty of games on those systems have subtle or unnoticeable affine artefacts.

>>2330224
>Early 3D requires less storage space than 2D. Vertices + textures are less data dense than everything is a sprite
Textures generally require far more storage than geometry, true on the N64 as well. In extremely limited space, variety or detail or both of textures is constrained. Early 3D only *requires* less space if textures aren't being used at all.

>Unfiltered only looks better for stylization... Filtered is objectively better for everything else.
Must I post it again? There's nothing special about bilinear filtering- it's a way of magnifying textures without aliasing, at the cost (or potential opportunity) of blurring if the textures are highly magnified.

>Let me guess, you're one from the emulator crowd that saw how badly N64 emulators filter textures and thought "it must be like that on the real console!". Well...it's not.
Huh? In your terms, it's objectively inferior on N64, a streaky 3-sample approximation of bilinear filtering. Of course, real bilinear filtering looks incorrect because N64 games weren't designed around, but it's a weird thing to bring up considering your argument.

PSX/N64 debate = swimmy console vs blurry console in a beauty contest to see who's the shiniest piece of shit in an aesthetically disastrous generation

>>2135450
#1

>I was simply pointing out what would happen if the nearest-neighbor upscaling was as botched as the upscaling of your bilinear example. No game engine produces the same effect as your swastika images, these are artifacts you've introduced through GIMP.
It's not botched and every engine would produce the same effect as my swastikas. Remember: as texels go to infinity, NN has no advantage over bilinear, because images are never magnified, always minified. The point is that with a finite approximation of a signal, it's strictly true that NN filtering can represent the signal better. If you have infinite texels, this is not the case; in the 5th gen, you had not merely "fewer than infinite" but fuck-all texels to play with.

See these Unity screenshots. On the left, the texture size is doubling from 8^2 up to 64^2, with bilinear filtering. On the right, the original 8x8 texture with NN filtering repeated over & over. I even put it at an angle to highlight NN's screen-space (as opposed to texture-space) aliasing. The left column is converging on the right column.

>it's usually part of a greater composite, say those nazi banners in Return to Castle Wolfenstein, which look far better when bilinear filtering is applied than nearest-neighbour.
RTCW had great enough texture detail (more than any 5th gen game) to pull it off (assuming the player never stands right next to a banner & looks at it). Those banners would look even better with bilinear filtered round parts & NN straight parts, antialiased with one of one of the modern full-screen post-processing AA algos (FXAA et al). Or even an alpha-cutout swastika with alpha-to-coverage. It's immaterial as 5th gen devs had access to neither technique. I do lots of graphics programming in my work and there are so, so many ways to skin a cat. Most were unavailable to 5th gen devs, as were detailed textures.

Bilinear inherently blurs things. That can be desirable or not.