Discrepancy in Cycles rendering result appears when "Pack Resources" on tga file #99565

New Issue

System Information
Operating system: Windows-10-10.0.19043-SP0 64 Bits
Graphics card: Quadro RTX 3000 with Max-Q Design/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 516.25

Blender Version
Broken: version: 3.3.0 Alpha, branch: master, commit date: 2022-07-08 16:07, hash: 2ee6891728
Worked: 3.2.1 release

Short description of error

The 3.3 Alpha version has a trouble to correctly render the tga texture, unless it is packed into the blend file.

After packing the external tga file, the rendering result becomes correct (see above).

Exact steps for others to reproduce the error
1, Open the sample file and load the tga texture;
2, File -> External Data -> Pack Resources;
3, File -> Save as;
4, Open the new saved file

The tga file and the sample Blender file are included.

Should not shine_3.blend

**System Information** Operating system: Windows-10-10.0.19043-SP0 64 Bits Graphics card: Quadro RTX 3000 with Max-Q Design/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 516.25 **Blender Version** Broken: version: 3.3.0 Alpha, branch: master, commit date: 2022-07-08 16:07, hash: `2ee6891728` Worked: 3.2.1 release **Short description of error** ![shine.jpg](https://archive.blender.org/developer/F13277668/shine.jpg) The 3.3 Alpha version has a trouble to correctly render the tga texture, unless it is packed into the blend file. ![not shine.jpg](https://archive.blender.org/developer/F13277671/not_shine.jpg) After packing the external tga file, the rendering result becomes correct (see above). **Exact steps for others to reproduce the error** 1, Open the sample file and load the tga texture; 2, File -> External Data -> Pack Resources; 3, File -> Save as; 4, Open the new saved file The tga file and the sample Blender file are included. ![piece.tga](https://archive.blender.org/developer/F13277655/piece.tga) [Should not shine_3.blend](https://archive.blender.org/developer/F13277659/Should_not_shine_3.blend)

Added subscriber: @Stay_Here

#101810 was marked as duplicate of this issue

Added subscriber: @EAW

Changed status from 'Needs Triage' to: 'Confirmed'

I can confirm that the issue occurs in 3.3 master hash: 443690604f
3.2.1 is ok.

Packed on the left, not packed on the right.

I can confirm that the issue occurs in 3.3 master hash: 443690604f4e 3.2.1 is ok. ![Packed vs unpacked.png](https://archive.blender.org/developer/F13283000/Packed_vs_unpacked.png) Packed on the left, not packed on the right.

Added subscriber: @LukasStockner

Caused by 33f5e8f239.

Caused by 33f5e8f239.

Added subscriber: @brecht

This is due to how OIIO's Targa loader is implemented.

In particular, it has an internal "alpha type" that it tries to read from the file. OIIO's internal alpha conversion logic applies whenever the file is 32bpp, it is not explicitly deactivated, and the file's alpha type is not explicitly set to "premultiplied". All three used to be the case for this file, until the commit linked above disabled OIIO conversion.

Instead, Cycles checks the oiio:UnassociatedAlpha spec return from OIIO now. Unfortunately, the Targa loader only sets this if the file is 32bpp and the alpha type is set to "useful". So, if the alpha type is neither "premultiplied" nor "useful", there is a change in behavior - the OIIO-internal conversion would have taken place, but the Cycles-side conversion won't.

See https:*github.com/OpenImageIO/oiio/blob/c2ed61bb853b4a446f74321c2fe4d665c01ef316/src/targa.imageio/targainput.cpp#L274 vs. https:*github.com/OpenImageIO/oiio/blob/c2ed61bb853b4a446f74321c2fe4d665c01ef316/src/targa.imageio/targainput.cpp#L837.

Apparently, the oiio:UnassociatedAlpha logic was changed here , but not the internal logic.

@brecht, since I see that you worked on that logic in the past - would you consider this an OIIO bug?

This is due to how OIIO's Targa loader is implemented. In particular, it has an internal "alpha type" that it tries to read from the file. OIIO's internal alpha conversion logic applies whenever the file is 32bpp, it is not explicitly deactivated, and the file's alpha type is not explicitly set to "premultiplied". All three used to be the case for this file, until the commit linked above disabled OIIO conversion. Instead, Cycles checks the `oiio:UnassociatedAlpha` spec return from OIIO now. Unfortunately, the Targa loader only sets this if the file is 32bpp and the alpha type is set to "useful". So, if the alpha type is neither "premultiplied" nor "useful", there is a change in behavior - the OIIO-internal conversion would have taken place, but the Cycles-side conversion won't. See https:*github.com/OpenImageIO/oiio/blob/c2ed61bb853b4a446f74321c2fe4d665c01ef316/src/targa.imageio/targainput.cpp#L274 vs. https:*github.com/OpenImageIO/oiio/blob/c2ed61bb853b4a446f74321c2fe4d665c01ef316/src/targa.imageio/targainput.cpp#L837. Apparently, the `oiio:UnassociatedAlpha` logic [was changed here ](https://github.com/OpenImageIO/oiio/commit/2086ed228bc265272988d61710f109191db7ca52), but not the internal logic. @brecht, since I see that you worked on that logic in the past - would you consider this an OIIO bug?

This looks like an OIIO bug to me, I would expect the internal logic and oiio:UnassociatedAlpha to match.

This looks like an OIIO bug to me, I would expect the internal logic and `oiio:UnassociatedAlpha` to match.

Added subscribers: @MeshVoid, @mano-wii

This issue was referenced by 94ec66c2f2

This issue was referenced by 94ec66c2f2859c717789a41cd5b8f0af0d7ec379

This issue was referenced by 0a35afbf86

This issue was referenced by 0a35afbf86c86963ebb638f0804a706d17df5b50

Changed status from 'Confirmed' to: 'Resolved'

I committed a workaround, though would still like to have a fix upstream as well. The issue is that it seems seems to have been an intentional change. Probably for files with the same alpha metadata as the files reported here, but different contents, and there may not be a single solution that handles all files correctly.

At least for us consistency with Blender image loading is most important.

I committed a workaround, though would still like to have a fix upstream as well. The issue is that it seems seems to have been an intentional change. Probably for files with the same alpha metadata as the files reported here, but different contents, and there may not be a single solution that handles all files correctly. At least for us consistency with Blender image loading is most important.

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Discrepancy in Cycles rendering result appears when "Pack Resources" on tga file #99565