We can rasterize a geometry into a layer. But how to do the opposite?

For example, say I have a rasterized layer of a teapot. Then I edit the layer's pixels.

Then how could I reproject the edited result according the the camera used for rasterization and the teapot's UV, and generate a layer that can be used as the teapot's texture?

Does the technique in time mark 1.36.00 here at "Houdini Hangout" help?

https://www.youtube.com/watch?v=hLa78ncQbTE&t=6154s [www.youtube.com]

This setup using a wrangle to 'reproject' is the best way I could figure out. I think there's probably a better way to transform world P to camera P that's built-in, but I don't know it. I'm using vex in this scene since it's what I know.

Thank you very much jsmack!

That being said, I feel it's really something should be built-in... there are so many nodes to bring data between 2D and 3D space, but not this?

By the way toNDC should be the built-in way to transfer world position to camera position. But somehow it returns clip space (?) not NDC so you need to divide by w yourself.

PHENOMDESIGN
Does the technique in time mark 1.36.00 here at "Houdini Hangout" help?

https://www.youtube.com/watch?v=hLa78ncQbTE&t=6154s [www.youtube.com]

Well I watched for a few minutes from 1:36:00... and I didn't find he showing any technique. He just added some nodes and deleted them. Do you mean this?



But it doesn't look like a reprojection from camera space to texture space either.

kodra
By the way toNDC should be the built-in way to transfer world position to camera position. But somehow it returns clip space (?) not NDC so you need to divide by w yourself.

toNDC was generating corrupt values for me, so I think it's not well defined here.

kodra
That being said, I feel it's really something should be built-in... there are so many nodes to bring data between 2D and 3D space, but not this?

I don't think there is a node, but I'm sure there is some openCL helper function for transforming the space.

jsmack

kodra
By the way toNDC should be the built-in way to transfer world position to camera position. But somehow it returns clip space (?) not NDC so you need to divide by w yourself.

toNDC was generating corrupt values for me, so I think it's not well defined here.

I've reported toNDC() issue to SideFX and they told me it's fixed... I guess not? I didn't try it myself again.

kodra

jsmack

kodra
By the way toNDC should be the built-in way to transfer world position to camera position. But somehow it returns clip space (?) not NDC so you need to divide by w yourself.

toNDC was generating corrupt values for me, so I think it's not well defined here.

I've reported toNDC() issue to SideFX and they told me it's fixed... I guess not? I didn't try it myself again.

It might be fixed in a newer version than I have installed.

Another thing is jsmack's solution retrieves camera information from SOP. Is it possible to somehow pipe the camera information into COP instead of referring to an SOP every time I need it?

https://www.sidefx.com/docs/houdini/nodes/cop/cameraimport.html [www.sidefx.com]

It seems like it's possible... but not sure how.

kodra
But it doesn't look like a reprojection from camera space to texture space either.

Sorry I did not catch the camera space, I confused for view space. You can try the Slap Comp that brings in a Solaris AOV. I imagine you can get a texture space readout from there?

https://www.sidefx.com/docs/houdini/copernicus/slap_comp.html [www.sidefx.com]

I saw the technique in Rohani Dalvi's video here: Houdini 20.5 - toon shading using copernicus [www.youtube.com]

It is not in SOPs though.