Image's Pixel method fails with 16 bit images #698
Description
Environment
- OS Version: Ubuntu 24.04
- Source or binary build? Either
Description
- Expected behavior: Images with 16 bit channels can be accessed
- Actual behavior: The Pixel function fails
Steps to reproduce
- Get a glTF with a 16 bit metallic roughness map
- Try to load it
Output
You will get this error printed out once for each image pixel (so 1 million times for a 1024x1024 map)
What is happening?
An RGB image with 16 bits per channel has a total of 48 bits per pixel (BPP in FreeImage terms). However, the implementation of FreeImage_GetPixelColor can only handle 8, 16 or 24 BPP, so it falls through the default case and fails.
I believe there are a few possible solutions to this:
Custom parsing for 48 BPP images
We could do custom parsing for 48 BPP (and I suppose 64 BPP, for RGBA 16 bit images) where we scan the image manually and get the pixel value.
Downsample to 8 bits
We could just downsample all images to 8 bits and accept that perhaps 16 bit maps are a bit overkill
Change library
We could change to a different library that can actually handle 16 bit images well, Freeimage seems to be pretty unmaintained (I thought there were some experiments in this direction?)
Do nothing and just fix the error message
At the very least we should print a warning for unsupported BPPs and exit early, rather than having a very puzzling pixel access error