I’m thinking about maybe getting a different SLR for photogrammetry but I don’t know the limit of my vram.
How many mp can work on 12gb vram.
You can find that out yourself.
Just take a handful of images, resize them to whatever you need, like 7664x3264 for a 25mp image.
For batch processing IrfanView or XnViewer can serve you well, but make sure they actually enlarge things. Normally you would not want to enlarge an image. But even an enlarged image would for sure tell you, if that resolution fits your VRAM.
Feed that to WebODM and watch the console output closely.
A few seconds after the start you will see a line like:
[WARNING] Image size (7664x3264px) would not fit in GPU memory, falling back to CPU
or if you are lucky a line that says:
[INFO] Photo dimensions for feature extraction: 4000px
[INFO] nvidia-smi detected
[INFO] Using GPU for extracting SIFT features
After a few seconds you can then already cancel the process.
I would also be curious to know 
Though, just to mention, feature-extraction via GPU is not a major improvement, neither quality nor overall processing time will see much change.
It’s more about the point cloud densification that your GPU will make a real difference.
And the densification process is not depending on VRAM at all.
Which means that your GPU with 12Gb VRAM will benefit you, independent of the resolution of your source material 
If you get around to try if 12Gb is enough for 25mp feature-extraction, I’d still be happy to know 
12gb is not enuf for future extraction with 20mp images.
Wow, that’s something.
I tried it with a card that has 6Gb VRAM. It renders 12MP fine, but even 12.5MP do not fit 6Gb anymore.
20MP not fitting 12Gb is quite a thing.
Nevertheless, luckily most GPUs can still handle the real work when doing the point cloud densification.
Yes the pointcloud generation really got fast.
1 Like
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.