Shadow casting is very expensive, most (almost all) dynamic lights do not cast shadows, same as in all games, nothing specific to Crysis 2.
Range of graphics effects in console games *spawn
- Thread starter Cyan
- Start date
Shadow casting is very expensive, most (almost all) dynamic lights do not cast shadows, same as in all games, nothing specific to Crysis 2.
That is actually my main gripe about the (over?)use of blur in video games.
I can understand the cinematic impact that it has as a powerful tool used by directors to focus the attention of the viewer on a specific part of the scene, but when it comes to actual gameplay I prefer having no blur at all (or perhaps just a very subtle effect).
In many games with blur effects I notice that when you look directly anywhere but the center of the screen, it sometimes just look wrong - as if your own eyes can't focus properly.
I'm not sure how accurate and resource hogging headtracking is right now, maybe in next gen games we'll have Kinect (or whatever comes next) tracking our eyes - and the game engine will adjust focus accordingly :smile:
Doesn't the main benefit of the imposter tech actually have to do with the fact that it's saving time for the artists? As far as I understand, it frees you from the need of creating different LOD models and basically automatically creates (on the fly?) multiple LOD models from the actual high poly model.
Impostors can be generated either off-line or on-the fly. It amounts to creating the impostor textures (storing a discrete set of different angles of the object) and applying them to a single polygon that itself doesn't change. Like sprites! [strike]From what I recall of Reach and examination via Theater mode, the impostors are not animated much if at all. [/strike]When their screen-space position changes, the texture atlas is consulted for loading up the different angle view; you can actually see a bit of this with the warthogs spread around on the ground in the opening scene of level 5 in Reach.
There's a bit of a data explosion with characters that are animated, which need different angles per animated frame. In the Mass Effect 2 scene, I believe the characters are mostly stationary with a few animations.
As for multiple LOD "models", it's possible yes, but you're still going to be limited to a discrete set of view angles so it's going to look bad with them being closer to the camera; it'll really depend on the scene though as to what is adequate. It's really multiple LOD impostor textures, not geometry. In both of those two games, the objects are quite far away from the player camera. Fidelity-wise, it can range from a rather simple colour shaded image to something with more texture detail, normal maps or different materials/lighting. The vehicles in the opening scene of level 5 in Reach were lit differently from one another and had more details with the cut-scene camera being placed relatively closer to them. This is in contrast to in-game where you'd generally have vehicle battles taking place kilometers away. Examination of the latter in the Theater mode revealed them to be rather simple, mostly just the general colour of the vehicle. With the theater camera being zoomed in and out, I don't believe they were generated in-real time, and were simply tailored for the scenes they were going to be used. Same with the Mass Effect 2 example as you can walk closer to the distant crowd with no change in the image detail.
At some point, there's just no replacing the 3D model, and perhaps tessellation-managed LOD is ideal.
edit: hopefully that's clear...
There's a bit of a data explosion with characters that are animated, which need different angles per animated frame. In the Mass Effect 2 scene, I believe the characters are mostly stationary with a few animations.
As for multiple LOD "models", it's possible yes, but you're still going to be limited to a discrete set of view angles so it's going to look bad with them being closer to the camera; it'll really depend on the scene though as to what is adequate. It's really multiple LOD impostor textures, not geometry. In both of those two games, the objects are quite far away from the player camera. Fidelity-wise, it can range from a rather simple colour shaded image to something with more texture detail, normal maps or different materials/lighting. The vehicles in the opening scene of level 5 in Reach were lit differently from one another and had more details with the cut-scene camera being placed relatively closer to them. This is in contrast to in-game where you'd generally have vehicle battles taking place kilometers away. Examination of the latter in the Theater mode revealed them to be rather simple, mostly just the general colour of the vehicle. With the theater camera being zoomed in and out, I don't believe they were generated in-real time, and were simply tailored for the scenes they were going to be used. Same with the Mass Effect 2 example as you can walk closer to the distant crowd with no change in the image detail.
At some point, there's just no replacing the 3D model, and perhaps tessellation-managed LOD is ideal.
edit: hopefully that's clear...
They can but dont have to. If they do they then multiple shadowmaps are needed according to amount of lightsources. Thats quite perfomance heavy. Pretty much all games have only a very small amount of lightsources that cast shadows per frame and most games only let 'tactically' placed lightsources cast shadows. That said I think Stalker games are only games I've seen with tons of shadowcasting lightsources per frame (muzzleflashes+other lights). Pretty much all lights cast shadows.
Tessellation can't fully solve LOD issues. In fact if you're using higher order surfaces, you'll generally need more geometry on the base (untesselated) control mesh compared to a highly optimized current day model.
For example, the KZ2 enemy characters have about 10.000 triangles, and each LOD halves the poly count, so they have 5000, 2500 and 1250 poly versions. But to build the same character with tessellation (and even displacement) you'll probably need more than 10000 polygons.
So even though you can then easily increase the geometry resolution for close-ups, you'll still not automatically get the low detail LODs just by turning off tessellation. You'll probably still have to build discrete LOD models separately, and then you'll still have to deal with the transitions for the low detail levels.
There are some very good automatic poly optimization tools that have been around for years, and one of the most commonly used programs, Zbrush, also has one integrated (although it's more suited to reduce very large polygon counts for sculpted models, and I haven't tried it for ingame stuff yet). They can even keep texture UV coordinates so very little manual work is required.
There are things I expect from UE3 to be improved and I'm very surprised that they still are lacking in that department.I'm thinking about their motion blur and depth of field implantation,especially the later.It looks like bloomed fog and not depth of field
I'm not sure which game you had in mind, but Gears 2's implementation is certainly not a bloomed fog during cut-scenes. During gameplay, it appears to reference the depth of the iron sights reticle point and blurs the scene beyond that according to the distance. This is especially apparent in the larger scale environments with a larger range of depth. For gameplay purposes, I don't see why one would want a heavy blur for the DOF in the immediate vicinity. There are more elaborate DOF filters, but in the end it's a trade-off, and the DOF as-is is inexpensive.
Even the new DOF being pimped in UDK is only meant for cut-scenes.
It probably makes a lot less sense if you have just a single zoom-level. In a game with iron sights, I can see it being of use for enhancing the location/distance of the targets as your FOV has just become extremely limited, and you are intentionally trying to aim at an enemy - it's in your control.
Of course, the depth/distance should be large enough for it to make sense.
It's an interesting thought! A couple things I'd worry about would be input lag and the variety in setups for end-users (size of the screen, sitting distance) and user preference between moving one's head vs one's eyes.
Does this mean that the Banshees, Ghosts, and other vehicles that are fighting in the background in certain Halo: Reach levels aren't "imposters"? Because while looking at them in Theater mode, they definitely seem polygonally based.
Ok, let's try this again:
It's possible they have all their angles covered.
They're pretty low res, so it shouldn't take much memory. Maybe they're generated on the fly so as to get the animation part (memory savings too). Who knows. I just saw that enemies at the beginning of Tip of the Spear will switch to a really low res asset whilst still being animated if you move the camera far enough. Zooming in returns them to the higher res model. (Take back that comment on not-animated...)
The banshees and falcons you're referring to could just be simple 3D models, but I mean, they're probably impostors...
edit:
Here's an example impostor texture with animation frames (shamelessly stolen from this) :
It's possible they have all their angles covered.
They're pretty low res, so it shouldn't take much memory. Maybe they're generated on the fly so as to get the animation part (memory savings too). Who knows. I just saw that enemies at the beginning of Tip of the Spear will switch to a really low res asset whilst still being animated if you move the camera far enough. Zooming in returns them to the higher res model. (Take back that comment on not-animated...)The banshees and falcons you're referring to could just be simple 3D models, but I mean, they're probably impostors...
edit:
Here's an example impostor texture with animation frames (shamelessly stolen from this) :
Memory consumption should be fairly reasonable, and for the various enemy ranks, they can just resort to using different colours (orange elite vs blue elite) whilst keeping one set of impostor textures in memory for elites. Wonder if memexport could be/was used...
I think so. Honestly, if the consoles pull off the Crysis 2 cutscenes at mediumish quality in real time I'll be extremely insanely off-the-charts impressed.
The perceived quality is much better though. With a gaussian type of blur you don't get the Bokeh spots, and Crytek makes good use of them.
L. Scofield
Veteran
Well, it's a daylight scene. You won't see that much in terms of dynamic lights.
Doesn't seem to be expensive at all. Deactivating it doesn't seem to affect performance that much. Seems like Crytek found a very good way to optimize the effect.
corduroygt
Banned
How does the C2 (console) DOF compared to UC2 DOF? The latter was the best DOF effect I've seen so far. Will Chris Balestra have to eat his words when he said the quality of DOF they had wasn't possible on the 360?
uh........ ok? The question was about the number of dynamic lights that could cast shadow maps in general, not the mere existing of shadowmaps being used in the game. Obviously not all of the game is going to be just the daylight scene.
L. Scofield
Veteran
Of course, but then shouldn't we use footage from levels more suited for this type of analysis, like the E3 demo at night?
What did he mean with that statement?
