Starship Asterisk*

Nitpicker wrote:Show me an interpolated version of this particular object which looks better.

My criticism is in response to the savaging this APOD received from Chris and then you. I felt that your criticisms were wildly overstated. From one real frame to the next, there are so many parts of the dust cloud which show matching similarities in shape, but which are noticeably displaced in the plane of the image. The decision to interpolate may be questionable, but once made, I can't imagine a better interpolation method for this particular case, than the morphing method employed. Sure, the morphing does some strange things to the finest "structural" details of the cloud, but in terms of the bigger picture of being able to map the variable density of the dust cloud in 3-D, I don't imagine this approach is terribly flawed. Learning more about light echoes was one of the great aspects of this APOD. But learning of a light echo's ability to provide data to build 3-D spatial models, was even more interesting.

The point is that it should not be interpolated. Just because you can interpolate doesn't mean you should. There are exactly three frames that are close enough together to see clearly the light moving through the medium. The rest are ridiculously far from one another. I have a feeling the only way to really show it to you would be to turn back the clock and devote Hubble to at least a couple dozen more frames to put it all together. As for my savaging of this APOD, as you have put it, I only started to feel strongly about this after I noticed many people other than myself expressing great confusion after viewing the video.

Sorry that you appear to reject my point about the 3-D modelling aspect. I do understand yours and I don't reject it entirely, and you don't really need to show me anything additional and I don't think you were being anywhere near as savage as Chris. It is hardly the first time that a significant number of people have failed to understand an APOD. I am not convinced that a non-interpolated version of this particular object, would have been significantly better understood. Nor do I think that this particular object (in any presentation) is a great example for understanding what a light echo actually is. I recall that at the time, I had to go off and do some additional reading, before I really understood what this APOD was trying to tell me.

I actually have no idea what your 3d modelling thing is.

geckzilla wrote:I actually have no idea what your 3d modelling thing is.

Oh. Huh. You mean you can't understand how one could build a 3-D model of the dust cloud's density, from this data? It is a pretty rare thing to get a feel for the 3-D shape of a nebula. That was the thing that immediately grabbed me about this APOD and this object. And I think the inclusion of the morphed frames (mapped onto ever-expanding ellipsoidal surfaces in 3-D) would probably make for a more realistic 3-D model, than if they were not included.

I get that you could build a 3d model but I don't think you would want to use this kind of interpolation to fill in the gaps. The 3d model would be pretty crap, anyway. You'd be better off modelling a completely made up sphere of dust with similar features and density derived from the real data.

geckzilla wrote:I get that you could build a 3d model but I don't think you would want to use this kind of interpolation to fill in the gaps. The 3d model would be pretty crap, anyway. You'd be better off modelling a completely made up sphere of dust with similar features and density derived from the real data.

That's rubbish. What I think you are describing would be yet another kind of interpolation, more convoluted than the interpolation used in the APOD.

It would be a model. There's no point in trying to replicate the thing accurately when any attempt would be so inaccurate that it would be useless anyway. If you think it's easy to tell what's in the foreground and what's in the background then I think you are mistaken. The best you can do is model small parts which happen to be nearly (or appear to be) perpendicular to our view.

geckzilla wrote:It would be a model. There's no point in trying to replicate the thing accurately when any attempt would be so inaccurate that it would be useless anyway. If you think it's easy to tell what's in the foreground and what's in the background then I think you are mistaken. The best you can do is model small parts which happen to be nearly (or appear to be) perpendicular to our view.

Yes, I do think it would be easy to make a good, first-order estimation of the density in 3-D, so we would appear to be in disagreement on that point. I think brightness from secondary and higher echoes would largely be negligible. Looking at the sequence of real frames, it appears that the unrelated foreground/background stars get brighter. This suggests a longer exposure was required for each successive image, to adequately expose each successive primary echo shell. This makes sense as the amount of dust between the camera and the primary shell, increases with each frame, and also just because of the inverse square law. (I don't understand your last sentence.)

Nitpicker wrote:Sorry that you appear to reject my point about the 3-D modelling aspect. I do understand yours and I don't reject it entirely, and you don't really need to show me anything additional and I don't think you were being anywhere near as savage as Chris.

Savage? Really?

All I said is that I don't like this movie because it significantly distorts reality. Because it is confusing and misleading. I've seen the animation made with just the actual data, and it is much clearer. Can it be misunderstood? Certainly. The first time I saw it I had to work to see it as a moving pattern of illumination and not as moving material. But I could do that, by careful examination. It is impossible to do that with the morphed image, because it actually relocates imagery.

Many image processing techniques result in artifacts. Sometimes it's the price we pay for getting clarity in some other area. That isn't the case here. This processing adds no useful information at all, but does obfuscate what is actually occurring. That's bad.

This is not something you would use to construct a 3D model. If you want to do a paper search, you'll find that people have used light echo imagery to deduce 3D structure. They use tomographic techniques. They do not work with interpolated frames at all, and certainly not with ones that are interpolated using standard morphing techniques, which only work for moving material, and only when we can reliably map a point in one frame to a point in another (which is a concept that makes no physical sense with light echos).

These are nothing but simple observations. To call this savage or vicious is more than a little over the top.

Chris Peterson wrote:That's because this is a terrible animation (and, IMO, a terrible choice for an APOD).

Yeah, I'd call that savage and overstated.

What I'm describing is a tomographic technique and if you've got massive gaps in your data, your only options are to interpolate or just accept the gaps. And I can quite easily, and I think reliably match several regions in each real frame, to several regions in the next and I fail to see how it makes no physical sense. I cannot think of a better interpolation method to suit a 3-D model of this data.

Nitpicker wrote:(I don't understand your last sentence.)

I meant that only a 2d disk centrally crossing the light sphere perpendicular to our viewing axis could be somewhat accurately mapped.

Nitpicker wrote:

Chris Peterson wrote:That's because this is a terrible animation (and, IMO, a terrible choice for an APOD).

Yeah, I'd call that savage and overstated. :ssmile:

What I'm describing is a tomographic technique and if you've got massive gaps in your data, your only options are to interpolate or just accept the gaps.

Nope, you are not describing a tomographic technique, and you do not interpolate 3D structure by creating interpolated spatial imagery from the raw data. You use the math to build the 3D structure from the real data, and then you can consider interpolating in that output space.

Again, look up some papers about how light echos are used to calculate possible 3D structures.

Chris Peterson wrote:

Nitpicker wrote:

Chris Peterson wrote:That's because this is a terrible animation (and, IMO, a terrible choice for an APOD).

Yeah, I'd call that savage and overstated.

What I'm describing is a tomographic technique and if you've got massive gaps in your data, your only options are to interpolate or just accept the gaps.

Nope, you are not describing a tomographic technique, and you do not interpolate 3D structure by creating interpolated spatial imagery from the raw data. You use the math to build the 3D structure from the real data, and then you can consider interpolating in that output space.

Again, look up some papers about how light echos are used to calculate possible 3D structures.

Then you don't understand what I'm describing.

Nitpicker wrote:Then you don't understand what I'm describing.

Maybe not. But I do understand tomographic reconstruction, and you are certainly not describing that.

And true, I probably wouldn't use "morphing software" to interpolate this data in 2-D, for the purpose of making a 3-D model [if it were my job]. I would identify related regions from each real 2-D frame (ellipsoidal shell in 3-D) and fit a curve to them in 3-D, to form the basis of a non-linear 3-D interpolation. But I imagine that "morphing software" provides a not altogether terrible approximation to this, at least for the purposes of an animation of the observations. Granted, it does false things to the detailed structures, but if you view it more as a density map, it suddenly looks less bogus.

Chris Peterson wrote:

Nitpicker wrote:Then you don't understand what I'm describing.

Maybe not. But I do understand tomographic reconstruction, and you are certainly not describing that.

Well, I suppose I haven't actually described it in much detail, but that is what I have been implying.

Sorry that it has taken me so long to properly clarify my reasoning as to exactly why I think the morphing is not terrible. From the outset, I have been viewing this APOD as a 3-D density map of the dust cloud (or a means to that end) rather than as an explanation of a light echo. I rarely pay close attention to the fine details of nebulae at the best of times, but such details were especially unimportant to me in this case.

Morphing software (which is too generic a term anyway) is not merely intended to simulate material movement (indeed, I'm not even sure if that is its most common application). It is intended to provide for smooth transitions from one true scene to the next, where there are identifiable similarities between each true scene. I doubt that any commenter here, could successfully identify the precise form of interpolation algorithm used in this APOD, in terms of the mathematics. But after studying the 2-D data and the 3-D geometry for a while, my educated guess is that the density map results of the algorithm applied to the APOD, are not likely to be terribly far from the results of the best possible non-linear, weighted, 3-D interpolation.

Nitpicker wrote:Morphing software (which is too generic a term anyway) is not merely intended to simulate material movement (indeed, I'm not even sure if that is its most common application). It is intended to provide for smooth transitions from one true scene to the next, where there are identifiable similarities between each true scene.

All morphing software I've ever seen operates the same way. It assumes structure moves. You map reference points in one frame to the relocated reference points in the next, and the software actually moves those pixels, and interpolates motion for all the other pixels around the reference points. This will fail with a light echo, just as it will fail if you were to use frames made of a room where nothing is moving but you change the position of the lighting. You'll get hideous artifacts as the software moves fixed structure around.

Chris Peterson wrote:

Nitpicker wrote:Morphing software (which is too generic a term anyway) is not merely intended to simulate material movement (indeed, I'm not even sure if that is its most common application). It is intended to provide for smooth transitions from one true scene to the next, where there are identifiable similarities between each true scene.

All morphing software I've ever seen operates the same way. It assumes structure moves. You map reference points in one frame to the relocated reference points in the next, and the software actually moves those pixels, and interpolates motion for all the other pixels around the reference points. This will fail with a light echo, just as it will fail if you were to use frames made of a room where nothing is moving but you change the position of the lighting. You'll get hideous artifacts as the software moves fixed structure around.

This is what the term "morphing software" means to me:
http://en.wikipedia.org/wiki/Morphing

And your example of a room is not a good comparison. Perhaps if there was an elephant in the room, you'd get a pretty good idea of the 3-D shape of the elephant, and you could perhaps overlook the fact that its ears were not quite right, or that the lamp in the corner looked a bit funny. Again, it is not about the fine details in this case, it is about 3-D mapping. Having said that, 3-D laser scanning technology for terrestrial subjects has come a long way in the last 20 years. I have seen a live demonstration produce quite a good, detailed, 3-D point cloud of a room I was sitting in (with no elephant), even though that is not its primary application.