I recently found an interesting post / document on how to do proper photogrammetry for VR/Unity.
I thought I’d share it as it seems to be well written and may help some of you improve your photogrammetry results.
The document can be found here:
I recently found an interesting post / document on how to do proper photogrammetry for VR/Unity.
I thought I’d share it as it seems to be well written and may help some of you improve your photogrammetry results.
The document can be found here:
Wow fantastic - last update only this morning! Half read it so far but:
“For ultra-wide lenses, only use rectilinear lenses, not fisheye lenses. Fisheye lenses are not recommended because they have too much distortion.” - is this particular to RC, which the author is using? I thought I read that RC is happy with fisheye.
Because this https://www.youtube.com/watch?v=8HuOvf4rKaw strongly prefers fisheye over wide rectilinear, principally using the ‘higher resolution’ (?) centre of each photo, the larger periphery as a sort of bonus for better context.
Hi ,
thanks for sharing this!
I browsed through it real quick.
What I find very useful are the graphs for properly capturing interiors - the bes I’ve seen so far.
However, there are many points in there with which I strongly disagree. I don’t want’ to just badmouth the article, but use it as a basis for discussion. After all, it provides a very concise overview about many important steps in the workflow.
The recommended RAW developement includes many processes that are usually big NoGos like rectification and de-noising. Although I do know that rectification is not necessarily as bad as some experts make it sound, I would certainly not do it in my RAW workflow because this will just result in an unneccessary loss in quality. De-noising is supposed to be just as bad - or does anybody have some good experience with it? What’s also usually not recommended is sharpening - you won’t get sharper edges by doing that anyway.
Also the recommendations for camera equipement are rather a bit over the top imho. The author names some of the most expensive bodies in their respective categories, which I find is a bit misleading. For one thing there are by far less expensive bodies that have comparable qualities (with respect to the price tag). And for another, I really think that for the purpose af a VR experience, the last iota of accuracy is not really neccessary - although that depends on the workflow and of course the expectations (if you want to be able to examine an ivy leaf in front of your nose and see the veins, then you need everything you can get). While it CAN certainly be quicker to use high resolution cameras, I am convinced that comparable results are possible with only a bit more effort but therefore much cheaper equipement, and the difference can be invested in other neccessities like software etc. This is a very important point for me especially for people who want to branch out into this field. At the beginning I was very apologetc for using my low grade (although well chosen) equipement. But now I know that an expert can achive much better results with crappy equipement than a beginner with the most expensive one (which is true for almost anything, mind). I would definitely recommend to beginners with a thight budget to either use a camera they already have or get a rather cheap but still suitable one (I use the LX100 for example) and then, after gaining some experience, move on to a more expensive one.
The big point of the author for chosing those high end bodies is the noise of the sensor, the quality of which is supposedly tied to the size of it. However, this also depends a lot on the quality of the sensor and the processing of the signals by the in-camera software (yes, RAWs are also processed, just usually not nearly as much as jpegs). A good source for judging sensor noise is here: https://www.dpreview.com/reviews/image-comparison/ I was quite surprised by the performance of some cameras (positively AND negatively).
Furthermore, the explanation of Image Quality for texturing is not correct. What the author describes it what everybody THINKS it SHOULD be (at least I did as well), but really it is only an average approximation. It get’s less accurate the more complex a scene is because then there are huge differences from close details to the background. 100% simply means that the ACTUAL resolution of the unwrap is 100% the same as the APPROXIMATED resolution and not the MAXIMUM achievable resolution. Especially for objects close to the camera, a much higher resolution should be possible.
Anyway, please correct me if I’m wrong!
A nicely detailed document - Thanks for linking.
The link is dead. Does anyone have a valid link?
Hi MI_CI,
I think it is possible to google something similar. We don’t have the mentioned link, but if you google for how to _do proper photogrammetry for VR/Unity _you will get quite a lot of results.