I can guess a few things that you’re going to say about this one and to some degree I’d agree with them 
… but I have been doing a lot of testing with fisheye lenses over the last year and find that there are some particular benefits to using them. The shooting strategy is somewhat different even to ultrawides and even a circular fisheye can be useful in certain situations.
e.g. https://sketchfab.com/uomdigitisation/folders/william-ricketts-sanctuary, http://files.digitisation.unimelb.edu.au/potree/pointclouds/ricketts-test2.html
Would love to be able to combine rapid capture with rapid processing.
OK so it’s mostly there. For anyone else reading this… Prior lens distortion should be set to unknown (unless you have known/approximate values to enter)…
Ben Kreunen wrote:
OK so it’s mostly there. For anyone else reading this… Prior lens distortion should be set to unknown (unless you have known/approximate values to enter)…
I don’t see such an option anywhere in the “alignment settings”
brennmat wrote:
Ben Kreunen wrote:
OK so it’s mostly there. For anyone else reading this… Prior lens distortion should be set to unknown (unless you have known/approximate values to enter)…
I don’t see such an option anywhere in the “alignment settings”
FYI for anyone else confused by this: these options only show up when you have the actual images selected in the pane.
Have you had any success with ultra wide circular fisheye’s with mask? In particular, we have some with prior known calibrations using Division model (I believe that this is the model used by OpenCV fisheye module).
In a nutshell, for a learner, what is the benefit of fisheye (apart from ultra-wide-area capture - a lot of ground covered in one click, but at loss of pixel density). I too have a fisheye on loan, with a special purpose in mind, due to try as soon as all these Xmas preparations allow!
the advantage to fisheye is just general coverage . however the calibration for lens becomes all that more critical as the geometric distortion can produce larger errors and in turn throw off triangulation. it’s significantly easier to produce accurate results with a less distorted lens.
I guess that’s because projection lines would have to be progressively curved, to a pre-determined distortion formula, the further from centrepoint of the photo?
I’ve been wondering whether the exagerated ‘look sideways’ effect of a fisheye would provide a one-hit almost cylindrical view of the four reveals (embrasures?) of a window opening, when shot very close to the glass line. Then moving the camera sideways and/or outward would create more ditto with usefully different parallaxes. Been trying that today - disappointing results.
I’ve been doing a few more experiments over the holidays and decided to give this one a bump. The difference between using a fisheye distortion model and brown4 for a full frame fisheye is very significant. While I don’t recommend using Gopros for photogrammetry there are a number of people who have them and will try and use them regardless… in particular people who also shoot 360 video who want use the cameras in a multi camera pole-mounted configuration to quickly capture a scene. This might possibly open up RC to a broader suer base(?).
At present I can get very good results with other software but without RC’s speed which makes it impractical.