Cycle cross-product node does not give true cross-product #61031

New Issue

Nathan Vasil · 2019-01-30T18:36:07+01:00

Nathan Vasil commented

2019-01-30 18:36:07 +01:00

Win 10, Nvidia 1070
Blender 2.79b release

Cycles's vector math/cross-product node does not provide the true cross-product; it appears to provide a normalized cross-product vector instead.

crossbug.blend
File provided has a plane emitting vector->color either from Cycle's cross-product node or a home-built cross-product node group. It shows the cross of (0.1, 0, 0) with (0, 0.1, 0). True cross-product of this vector is (0, 0, 0.01) (source: https://www.wolframalpha.com/input/?i=(0.1,+0,+0)x(0,+0.1,+0) ).

See https://developer.blender.org/rB448d143ad0e73a9d2c86f84def2b9f89b8332470 . There was an original inconsistency, and it appears that a version of the cross-product with built-in normalization was used instead of the true cross-product.

I encountered this issue when trying to implement axis-angle rotation in nodes. Vector normalization nodes exist and I would expect anyone that knows enough to use vector math nodes will know when they do and don't need to normalize their vectors.

Documentation link at https://docs.blender.org/manual/en/dev/render/cycles/nodes/types/converter/vector_math.html .

Win 10, Nvidia 1070 Blender 2.79b release Cycles's vector math/cross-product node does not provide the true cross-product; it appears to provide a normalized cross-product vector instead. [crossbug.blend](https://archive.blender.org/developer/F6455523/crossbug.blend) File provided has a plane emitting vector->color either from Cycle's cross-product node or a home-built cross-product node group. It shows the cross of (0.1, 0, 0) with (0, 0.1, 0). True cross-product of this vector is (0, 0, 0.01) (source: https://www.wolframalpha.com/input/?i=(0.1,+0,+0)x(0,+0.1,+0) ). See https://developer.blender.org/rB448d143ad0e73a9d2c86f84def2b9f89b8332470 . There was an original inconsistency, and it appears that a version of the cross-product with built-in normalization was used instead of the true cross-product. I encountered this issue when trying to implement axis-angle rotation in nodes. Vector normalization nodes exist and I would expect anyone that knows enough to use vector math nodes will know when they do and don't need to normalize their vectors. Documentation link at https://docs.blender.org/manual/en/dev/render/cycles/nodes/types/converter/vector_math.html .

Nathan Vasil commented

2019-01-30 18:36:07 +01:00

Added subscriber: @vasiln

Sebastian Parborg commented

2019-01-30 19:03:38 +01:00

Added subscribers: @brecht, @ZedDB

Sebastian Parborg commented

2019-01-30 19:03:38 +01:00

@brecht bug or not?

Nathan Vasil commented

2019-01-30 22:52:09 +01:00

Not sure if you're asking me, but whether something's a bug or not is in the eye of the beholder. If you were to see cross3(vec3, vec3) and normalizedCross3(vec3, vec3) in some library, and both gave identical output, one or the other or both would be a bug. Even if they had identical code-- just on the basis of the name. Same as there's nothing buggy about a function named returnZeroNoMatterWhat(vec3), so long as it does what it says.

Both regular and normalized cross products are useful in vector math. All vector math nodes are sugar that can be created from scratch from more fundamental math nodes. Normalization is destructive in nature, and Blender has an existing vector normalization node, so my feeling is that there's no reason to provide a normalized cross in preference to a regular cross.

What I would say is that if you want a normalized cross, name it a normalized cross; if you don't want to name it that, document that you're using the term differently than it's used elsewhere. Otherwise, same as a coder using identical cross3() and normalizedCross3() from a library, some people will encounter issues and give up, other people will encounter issues and waste some time. But of course, since a cross product node is sugar to begin with, none of that is the end of the world.

That's my reasoning. Apologies if you weren't asking me and I misunderstood.

Not sure if you're asking me, but whether something's a bug or not is in the eye of the beholder. If you were to see cross3(vec3, vec3) and normalizedCross3(vec3, vec3) in some library, and both gave identical output, one or the other or both would be a bug. Even if they had identical code-- just on the basis of the name. Same as there's nothing buggy about a function named returnZeroNoMatterWhat(vec3), so long as it does what it says. Both regular and normalized cross products are useful in vector math. All vector math nodes are sugar that can be created from scratch from more fundamental math nodes. Normalization is destructive in nature, and Blender has an existing vector normalization node, so my feeling is that there's no reason to provide a normalized cross in preference to a regular cross. What I would say is that if you want a normalized cross, name it a normalized cross; if you don't want to name it that, document that you're using the term differently than it's used elsewhere. Otherwise, same as a coder using identical cross3() and normalizedCross3() from a library, some people will encounter issues and give up, other people will encounter issues and waste some time. But of course, since a cross product node is sugar to begin with, none of that is the end of the world. That's my reasoning. Apologies if you weren't asking me and I misunderstood.

Brecht Van Lommel commented

2019-01-31 17:15:50 +01:00

The naming is indeed poor, this should be fixed. A proper cross product could be added as well.

Nathan Vasil commented

2019-10-28 19:09:10 +01:00

Just wanted to add to this with some additional problems I've discovered with normalizing the cross vector.

The true cross product of parallel or antiparallel vectors is the 0 vector. When Blender normalizes a 0 vector, either internally in a cross-product operation or via a vector math/normalize node, it creates some kind of weird NAN/infinity vector. You can't do any kind of math on this vector, and it actually leaks through nodes you wouldn't expect it to-- for example, if you mixRGB with a fac of 0, from black to an infinity vector, you still get infinite values, even though the 0 fac should mean you get only black.

Again, the workaround here is to create your own cross-product node group out of float1 math, rather than using Blender's cross-product node.

I don't know if you'd consider this mixRGB thing to be a bug or not. If you need me to make a new bug report regarding it, just let me know.

Just wanted to add to this with some additional problems I've discovered with normalizing the cross vector. The true cross product of parallel or antiparallel vectors is the 0 vector. When Blender normalizes a 0 vector, either internally in a cross-product operation or via a vector math/normalize node, it creates some kind of weird NAN/infinity vector. You can't do any kind of math on this vector, and it actually leaks through nodes you wouldn't expect it to-- for example, if you mixRGB with a fac of 0, from black to an infinity vector, you still get infinite values, even though the 0 fac should mean you get only black. ![infinity.png](https://archive.blender.org/developer/F7866188/infinity.png) Again, the workaround here is to create your own cross-product node group out of float1 math, rather than using Blender's cross-product node. I don't know if you'd consider this mixRGB thing to be a bug or not. If you need me to make a new bug report regarding it, just let me know.

Marko Martinić commented

2019-12-27 21:54:46 +01:00

Added subscriber: @IndieBerrie

Marko Martinić commented

2019-12-27 21:54:46 +01:00

Changed status from 'Confirmed' to: 'Resolved'

Marko Martinić closed this issue

2019-12-27 21:54:46 +01:00

Marko Martinić self-assigned this 2019-12-27 21:54:46 +01:00

Marko Martinić commented

2019-12-27 21:54:46 +01:00

Fixed in https://developer.blender.org/D5523
In 2.81a, cross product is no longer normalized.
When you import old files that had cross product node, normalization node is added (in order not to break functionality of old files (you can always remove this added node))

Fixed in https://developer.blender.org/D5523 In 2.81a, cross product is no longer normalized. When you import old files that had cross product node, normalization node is added (in order not to break functionality of old files (you can always remove this added node))

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Cycle cross-product node does not give true cross-product #61031