motion blur (and animation)

[dougal2]

tomb,
he's not trying to render an animation with motion blur. he's trying to simulate very long exposures.

Imagine the case of photographing a car driving by at night with a very long exposure - the lights will leave a trail on the film that shows the path the car took as it went by. That path would almost certainly not be linear, or simple rotation about a point.

Currently, with lux there isn't a way to describe an entire motion path within a single frame, so he's merged all of the frames which make up that path.

[thomas]

hey, sorry I wasn't more clear about the intentions, but thanks dougal for explaining them.

This is the effect I was trying to achieve: http://archikey.com/buildings.php?id=35&picture=82
(ironically the roads are so straight in LA that in the case above it may have been possible to achieve that within a single frame , thus without the need for merging)

[tomb]

tomb,
he's not trying to render an animation with motion blur. he's trying to simulate very long exposures.

Imagine the case of photographing a car driving by at night with a very long exposure - the lights will leave a trail on the film that shows the path the car took as it went by. That path would almost certainly not be linear, or simple rotation about a point.

Currently, with lux there isn't a way to describe an entire motion path within a single frame, so he's merged all of the frames which make up that path.

Ah I see

Well, I still think using luxmerge is probably not the best way to go - perhaps best as postpro using a dedicated compositor. But LordCrc is probably the best person to comment on this
Using luxmerge I think you would need much much more than 100 frames to "sample" the motion enough.

T

[Lord Crc]

The correct term is motion blur. The object/frame is blurred due to motion while the shutter is open. The concept doesn't change just because the shutter is open 1s instead of 1ms

As dougal says, luxrender doesn't handle complicated motions during exposure. As such I think luxmerger could be a viable alternative. In this case, it is crucial that each frame is rendered with the same number of samples/pixel as all the others (otherwise noise will be unevenly distributed). Steps:

Set up a haltspp count (could be fairly low if you render a large number of frames)
Export the animation and render. Contrary to normal animations, you should NOT use the "fixed seed" parameter during rendering.
Finally use luxmerger to merge all the frames. If you're running low on disk space, you could merge the flm's after each frame, ie "luxmerger -o master.flm frame0102.flm master.flm".

Rendering and merging may be best handled by a small script.

If you're using a high number of frames (for something like your initial picture, I'd try around 500-1000 frames), you should be able to set haltspp to a very low number. I'd try 10 first. If you're not satisfied, just render the animation again and merge the resulting flm's from the two runs. If you wish to render the second run at a higher haltspp, you can do so, however just make sure ALL the frames in the second run are rendered with the new haltspp.

[dougal2]

Here's the result of a quick 10 'frame' non-linear motion blur using the technique described in this thread.

Obviously the path is not smooth because of the low number of 'frames', and the light source is brighter on the right hand side (towards the end of the path) because it was slowing down.

[thomas]

ok, that is quite a decent result... the overall path of the light source is much more clearly defined than in my first try.

The 'cloudiness' of the path is probably due to undersampling and shows that the spp must be actually quite high to get the high intensity pixels to have enough samples, or would maybe a high mlt strength be a good idea? Speaking of render settings what sampler and integrator did you use?

And what is the shape of your lighsource, an icosphere?
I ask becuase in my scene the number of cars is high and they have very detailed emitters. Say 50 cars x 300 faces = 15000 light sources. Could this high number of emitting faces result in that with a low spp some of the lights are missed completely when using path and thus result in the interrupted light path showing up in my render?

(my render:

)

[dougal2]

My emitter was just a cube. I used the 1st animation rendering preset - low quality, just 16 spp.
Yes, this image is massively undersampled, just for the sake of demonstration. It took perhaps just over minute to render all 10 frames.

[Lord Crc]

Indeed, your light sources may not be sampled at each frame. You could force it to sample all light sources (set strategy parameter to "all"), but that would take quite a lot of time I suspect. I think I'd try bidir with metropolis, but not setting the strength too high (0.4-0.5 would be my first guess).

[jeanphi]

Hi,

Your high number of lights will make the sampling quite difficult for sure, however there's no easy way to overcome this in a global illumination unbiased renderer (even biased techniques would have a hard time with such a high number of emitters I guess).
I have pending changes meant to improve time sampling for motion blur, I'll release them after v0.6.

Jeanphi

[thomas]

ok thanks for all your explanations, but best is probably to let this rest as you're priorities lay elsewhere atm and I have no priorities with this scene either, just investigated this out of curiosity.

Atm lux has a tough time sampling the lights in the scene, I did clean up the meshes but that didn't improve things. I just can't seem to get indirect samples from the motion blurred lights, you notice that when looking at my rendering posted above, the road isn't illuminated at all by the cars. I've tested all samplers and integrators all seem only to sample the direct path from eye to light and even for that it takes a long time for them to show up, hence the interupted light strokes in my rendering posted above. With bidirectional I can't get any illumination at all from the motion blurred lights. For reference I post an image rendered with and without motion blur (the latter without sunsky) and you'll see the point clearly.

bidirectional, without motion blur

path, with motion blur, without sunsky contributions

lightsonly.jpg (5.66 KiB) Viewed 1569 times

I do think it is something we should pick up at a given time in the future as a scene like this imo it really proves the point of continuously sampled motion blur.

If someone wants to have a go with it, I can post the scene somewhere but it is somewhat large.