" All of this effort, and it looks nothing like real camcorder footage."
Well, who decided that was the benchmark and explicitly stated goal of real time synthetic image generation? I've worked in the video games for several years, and I've never heard an Art Director/Artists tell us : "Make it look like it's shot by a camcorder" ... Of course some games do it, and do it fairly well.
The whole idea is to give the player an immersive visual experience, not to make you sea sick by watching wobbly footage from your uncle Ron's camcorder. You can achieve very realistic and immersive visuals without anything looking like a camcorder of some other type of video footage.
To clarify, I wasn't referring to animation. A camcorder is the only instrument that can produce a series of still frames that look identical to what a human observer saw in the same spot.
All simulated video falls short of that goal. But a camcorder can! That's why it seems like if we throw out our physics simulations and try to mimic a camcorder's output, we could do the same thing in software. Our renderer would look real by definition -- it'd be considered a bug for the renderer's output to deviate too far from whatever the camcorder would've recorded in the same orientation in real life.
I'm not saying it's a worthwhile goal, just that no one has achieved it yet. This seems like a possible strategy.
If anyone knows any prior research along these lines, I'd appreciate the references. It's pretty common to use a camcorder to tune an art pipeline, but the proposal here is to use the camcorder to force every single aspect of all rendering stages to produce a final output that looks identical to whatever the camcorder would capture in real life.
It's the total absence of artistic freedom: you wouldn't even be able to move the lights around in the simulation during the training stage, since the training corpus is a real-world scene with corresponding lights. But afterwards it (hopefully) would be able to extrapolate what other scenes should look like, just like an experienced painter can.
> A camcorder is the only instrument that can produce a series of still frames that look identical to what a human observer saw in the same spot.
Like a camera? A non-colourblind human with perfect 20/20 vision and no incidental visual impairment?
What is with the repeated camcorder mentions? A camcorder will make people think of something like the Sony Video8 Camcorder from the 1990s, or other early consumer-grade videocameras.
The video is the important part. If you say camera, people will think still frames. That's poisonous to the goal of realism, since video is the area no one has yet achieved.
Also, a camcorder from the 1990's still generates images that are far beyond our current capabilities of realism.
Video is still frames… you can't tell the difference between a 2-frame 24fps video and an animation of two photos taken 33ms apart. This holds as you add more images.
The things that record video for films and TV are called 'video cameras'. In computer graphics, the software concept representing the point of view for the rendered scene is called a camera. Calling these things cameras is standard and well-understood in this domain.
The obsession with the particular recording device seems like pointless minutia. If you called it 'a video capturing device', your message would not be weakened. Specifying exactly the model of 1960s triple-lens Bolex which the video has to emulate distracts from your message.
> That's poisonous to the goal of realism, since video is the area no one has yet achieved.
I don't know what to say: this has already been achieved, and you missed it. It doesn't seem important to expound on whether this has been achieved or not, since it's trivial to find an example which fulfills whatever your latest requirement is. cue "but realistic Hi
I still don't know what the message is you're trying to convey: real-time generation of convincingly near-photorealistic video will… something? Don't worry about correcting my definitions; that makes your message "I'm a specialised dictionary in human form*".
Well, who decided that was the benchmark and explicitly stated goal of real time synthetic image generation? I've worked in the video games for several years, and I've never heard an Art Director/Artists tell us : "Make it look like it's shot by a camcorder" ... Of course some games do it, and do it fairly well.
The whole idea is to give the player an immersive visual experience, not to make you sea sick by watching wobbly footage from your uncle Ron's camcorder. You can achieve very realistic and immersive visuals without anything looking like a camcorder of some other type of video footage.