Segfault when using relight-cli -L parameter

[ruven]

I keep getting segfaults when using relight-cli's -L parameter.

I'd like to be able to generate an image corresponding to an arbitrary x,y,z light position. I've tried with various configurations and what I presume to be the correct parameters: -D test.jpg -L 1:0.5:1 which should give an image with lighting from the middle right (is that how x,y,z work?). However, no matter what parameters I give I get a segfault. Does this functionality in fact work? If so, how should it be used?

(I'm using the latest Github commit and compiling on Linux)

[ponchio]

The -D parameter was ignored using instead the output parameter... sorry, I fixed it in the latest commit.

relight-cli -D test.jpg -L 1:0.5:0.5 rtifolder/info.json

This should workd ("/info.json" is optional).

Notice how x, y and z are supposed to be a directions, (a normalized unit vector), if it's not it get normalized, for example your parameter 1:0.5:1 becames actually 0.666:0.333:0.666 (indeed middle right a bit upper direction).

[ruven]

Thanks! It no longer crashes, but it still doesn't seem to be working properly. So, if I've understood correctly an xyz vector of 0:0:1 would be a light source pointing down from directly above and 1:0:0 would be raking from the right? No matter what vector I give, I only get a black image.

Also, I notice that you cannot save to JPEG - only PNG seems to work.

[ponchio]

Strange, I made a test using the linux portable version and it works properly (and can save both jpg and png) on a sample dataset (HSH).

Could you share dataset (with exact command line)? It could be a dataset problem, but it might also be a library problem (libjpeg and in general QT dependencies).

[ruven]

Ah, it looks like an issue with the type of basis function used to create the RTI. I just tried with HSH and it works as it should. RBF, however, doesn't work.

Try this example, which I created using RBF: relight-cli -b rbf -p 9 -q 50 and generated the reconstruction using relight-cli -D raking.png -L 0:0:1 ./info.json from within the folder.

test.zip

There's also possibly a QT issue somewhere. I tried on another machine running Ubuntu 22.04 (the first machine was Ubuntu 20.04) and this time was also able to export as JPEG.

[ponchio]

I left a stupid debug value in the code. Should work now. About QT: it migh have something to do with libjpeg version installed with QT and system or, maybe more likely, just some path missing: QT looks for a specific plugin folder where its libjpeg is supposed to be found. I am working in 20.04.

[ruven]

Thanks, it now works for RBF!

Can you confirm exactly how the -L parameter works? If I've understood correctly, the vector is the direction towards the light source, right? So 0:0:1 is a light from directly above? 1:0:1 would be a light at 45deg elevation from the middle? And 0:1:0.2 would be a low raking light from the top of the object? Is that correct?

I also tested with PTM format (relight converted from an old PTM file), but I'm getting color casting, so there's maybe another debug value somewhere there too!

[ponchio]

You are qualitatively right on the light directions., but you should work with unit length (x^2 + y^2 + z^2 = 1) vectors on the surface of an emisphere (z > 0).

Send me the sample and I will debug the PTM conversion problem....

On Mon, Jun 27, 2022 at 12:47 PM Ruven @.***> wrote:

Thanks, it now works for RBF!

Can you confirm exactly how the -L parameter works? If I've understood correctly, the vector is the direction towards the light source, right? So 0:0:1 is a light from directly above? 1:0:1 would be a light at 45deg elevation from the middle? And 0:1:0.2 would be a low raking light from the top of the object? Is that correct?

I also tested with PTM format (relight converted from an old PTM file), but I'm getting color casting, so there's maybe another debug value somewhere there too!

— Reply to this email directly, view it on GitHub https://github.com/cnr-isti-vclab/relight/issues/47#issuecomment-1167194723, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAUDV2YUNTSF4MD54ACXYODVRGBDNANCNFSM5ZUQMUYA . You are receiving this because you commented.Message ID: @.***>