sfs feature suggestion - variable blending per image

[Matt Hutton]

Hi,

In sfs the options blending-dist, min-blend-size and blending-power are specified once and used for all the images. I think that the possibility to set these values differently per image (just as one can with shadow-thresholds) would be useful, for example in cases where shadows are cast across some images from distant terrain and in cases where the images have different resolutions.

If there is some preferable alternative to this, or I have revealed I am going about things the wrong way, I would be pleased to hear this too.

Many thanks,
Matt

[Oleg Alexandrov]

Thank you for the suggestions. Yes, for different resolutions one would need different blending distances. 

It is not clear however how SfS would perform if images with various resolutions are used, unless the resolutions are not too different. The current approach does some rather naive bilinear interpolation from the ground to the images, and it is assumed that the DEM itself and the images are roughly at the same resolution.

Hope the tool is working for you. The registration part is always the hardest, and it appears that you are looking at advanced features, so hopefully all is good up to that stage. 

I would not be able to add more features for the time being. This year we had no funding for this project and even as it is I took some time from other projects (and sometimes free time) to address issues in it, so I hope perhaps with some careful tweaking of existing params you may arrive at some satisfactory solution. One more knob to tweak, btw, is the --robust-threshold. It prevents images which are having a hard time conforming to the solution from dominating it too much.

Happy to hear more about your progress. You can always email me apart from the mailing list with questions or issues (with the caveat that I have time for questions but not so much for adding new functionality). 

[Matt Hutton]

Thank you very much for these notes, Oleg. Might you be able to quantify how similar the image resolutions should be - would you expect a factor of 2 (say) to be OK? What about the relative resolution between images and DEM?

Yes, the tool is working very well, thanks. Once I realised I needed to use --mapprojected-data in bundle_adjust, registration was no problem.

Thanks for the suggestion of --robust-threshold, I'll certainly try that out and keep twiddling dials.

Matt

[Oleg Alexandrov]

Matt,

If the resolution of images and DEM for SfS are within a factor of 2, it should be fine. This is just a question of how well interpolation works, and usually its quality degrades gracefully as one samples an image either too finely or too coarsely. A bigger concern here is if the illumination is diverse enough among the various images, as that can make a big difference in the quality of the result. 

--
You received this message because you are subscribed to a topic in the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ames-stereo-pipeline-support/zkRN4nIFHIo/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/449d1d2b-6ea9-4e3c-8b8b-66403fba1870n%40googlegroups.com.