What would you like to know about photogrammetry?

polycounter lvl 4
Offline / Send Message
Pinned
Sebvhe polycounter lvl 4
Hi guys!

About a year ago I made a tutorial about photogrammetry (the one you can get in my signature). Since then it has been downloaded countless times, and I'm super happy about that!
At the time it did mostly focus on tileable textures, which I used to do a lot.
However it has been more than a year and I have learned so much since, I think it needs to be updated (not that it is not accurate anymore, but it could be improved). Rather than just update I feel like it would be better to make a completely new tutorial, that doesn't focus so heavily on textures, but more on photogrammetry for games in general. I started laying down everything that I could include and quickly realised that if I want to be "in-depth" it could have hundreds of pages.

Therefore I'd like to know more about what you guys feel would be the most useful :smile:

  1. Would you rather get a not-too-long but not-too-in-depth tutorial, or do you not mind having hundred of pages to read? (I add many pictures don't worry :wink: )
  2. What do you think is under-explained in the photogrammetry field so far? Capturing pictures? Processing pictures? Cleaning cloud? De-lighting? Converting to PBR? etc...
  3. What would be the reason you'd like to use photogrammetry the most? Rocks? Foliage? Buildings? Textures? etc...
  4. What do you think is the biggest issue with photogrammetry at the moment? What could be improved?
  5. If you guys read my previous tutorial, what was lacking? On the other hand, what was especially great?
  6. What do you think is the best format for such a tutorial? I usually like simple pdf the best, as it is easy to go back and forth, but wouldn't be against a video tutorial if you guys think it suits best. Also, do you think it needs projects files alongside it to help reproduce the steps?
  7. What about a website/blog that assesses small steps/subjects at the time in order to keep the tutorial more alive and up to date?

That would really help me narrowing down people's needs :smile:

Last thing, I'm a strong believer that education should always be free, and I'm never going to ask any money for tutorials. I do this because I like photogrammetry a lot and would love to see more people doing it!

Replies

  • chrisradsby
    Offline / Send Message
    chrisradsby polycounter lvl 8
    Capturing reflective metallic surfaces seems kind-of under-explained, I guess capturing bigger objects is always interesting, whole cliffs or big trees.

    I think PDFs work fine overall :)

    Nice work by the way , really appreciate it :)!
  • yabestpal
    1. I would prefer a shorter tutorial rather than 100's of pages to start personally. I think that many pages would get a little overwhelming, unless you maybe separated those 100's of pages out into different tuts teaching different concepts.

    2. I'd definitely be interested in learning how to convert textures into pbr, but I haven't dived in enough to photogrammetry really know what the most under explained concepts are.

    3. I would probably mostly use it for textures but being able to photoscan larger objects like cars and statues would be good to know.

    Pdf's are always welcome but I'm not opposed to there being a site either, also I wouldn't really need a project file for each step I think. When I do tutorials I usually open the final file to see how it compares to mine instead of opening one up for each step.

    Also thank you for doing this! Looking forward to reading your tuts!  B) 
  • Sebvhe
    Offline / Send Message
    Sebvhe polycounter lvl 4
    @chrisradsby : That's really interresting, I never really approached reflective/metallic surfaces. I know they can be scanned using some tricks. Thanks for putting that on my list :smile:

    @yabestpal : Thanks for the good feedback! How would you feel about having first a quick tutorial, and then afterwards some more in-depth tips and tricks? I don't really know how I could articulate this, but it may be more straightforward for the reader if done well.

    As for the bigger objects, it would really depends on the subject. As much as I would love to cover drone scanning and stuff like that. I first have to buy a drone and also feel it would only be of use to a tiny fraction of the people doing photogrammetry (for obvious reasons)
    I really want to focus on a tutorial that requires little to no gear and still brings great results.
    That being said I will definitely have a more in-depth look at big objects.

    Furthermore, since I wrote my first tutorial, the photogrammetry industry has changed quite a bit. More and more people are moving from Agisoft to Reality Capture. Reality Capture has undoubtely outclassed Agisoft, in both quality and speed. I'm therefore considering using Reality Capture for my next tutorial. Is this something that matters to you or would you be fine with any software?
  • littleclaude
    Offline / Send Message
    littleclaude polycounter lvl 8
  • yabestpal
    Offline / Send Message
    yabestpal vertex
    @Sebvhe I wouldn't have any problem with a quick tutorial followed by tips and tricks, whatever will get me making those sweet sweet textures and models fastest.

    And with that it sounds like moving the tutorial to Reality Capture just seems like the better move, Since I haven't actually used photogrammetry just interested in learning it the software choice doesn't matter too much to me from the get go. Although I imagine there are some Agisoft users out there that might need some convincing to make the switch. It might be cool to see the same object/texture captured uniquely by each software so we can see why one software is better than the other in terms of object accuracy/texture quality.
  • musashidan
    Offline / Send Message
    musashidan sublime tool
    Yes, I was going to ask that. There are so many software options available now and I've only tried PGram with Agisoft a few years back. I have a much better camera now and was thinking of revisiting so a reboot tutorial with Reality Capture would be most welcome.
  • Scruples
    Offline / Send Message
    Scruples polycounter lvl 6
    I'd like to know how google got such nice trees using only photogrammetry and what looks like 80% coverage. Is it the computer automatically replacing what it identifies as trees with what I guess would be just a sparse model of a tree/s?
    edit: oh! I should mention that's a screen from this awesome vidya of how google does their photogrammetry, it's mostly for plebs and doesn't go into detail though.

  • John Baxter
    Offline / Send Message
    John Baxter polycounter lvl 7
    I wish we could download Google's photogrammetry data in the form of textured meshes. The detail is pretty phenomenal considering the scale.
  • AceOfQuades
    I've been doing a lot of scanning the past few weeks and understanding what settings to tweak and prioritize when getting correct exposure is an under-explained area that I ended up working out through trial and error. If your new tutorial went over that I think it would help out alot of people. I'll add to the mix my experience that cranking ISO is something people shouldn't be afraid of. I was getting shots early on where I would take my shutter speed and F-stop as low as they can reasonably go for my subjects but keeping the ISO at 200. This was a heavily occluded area that just wasnt getting much light. Later on the darkened regions weren't exposed properly so all that came out was useless noise that was artificially recovered. Later in a similarly occluded location I did the same thing, dropping the other settings first, but still the subject was under-exposed. So I cranked the ISO up to 800. While this increased over-all noise, it meant that the camera was still able to capture actual light information from darkened areas which meant the picture was clearer for agisoft to read and find patterns.
  • Sebvhe
    Offline / Send Message
    Sebvhe polycounter lvl 4
    @yabestpal : there are tons of comparisons between Agisoft and RC and others solutions on the web. I personally think that both Agisoft and RC deliver really decent results. However RC seems to be able to handle many more pictures while being much faster. Agisoft on the other hand can get really frustrating because you'll quickly be limited by the amount of RAM your computer has.

    @Scruples : That video matches my understanding of it. Basically it's not much different than when I create my textures. When you scan surfaces (i.e. not 360 full meshes), top down overlapping pictures is by far what delivers the best results. The main difference here, is they have got a freaking plane, several super high end cameras and some serious processing power.
    Also I do believe their photogrammetry software can somehow understand what is a building, what is a tree, what is terrain. The corners are very very sharp on the buildings. They probably change the reconstruction aggressiveness according to content.
    And I don't think they replace trees with a library. The trees from my home look way too accurate to be generic, also, when multiple trees are close together they form one envelope rather than intersecting meshes.

    @AceOfQuades : Thanks for your input. Although camera settings isn't very complicated, it is somehow very, very hard to properly explain, and does require some experience to get right. I did cover it in my previous tutorial, but I only scraped the surface. I'll make sure it's crystal clear next time :smile:
    I am very afraid of cranking the ISO and it is something I used to always warn people of. However I recently came to realise that it is because my camera specifically (which is a cheap entry level DSLR) deals very poorly with noise.
    Having tried a few other DSLR you could definitely crank the ISO to at least 400. I would totally follow your position about it on my next tutorial.

    Your feedback has been invaluable so far, I definitely didn't think about most of the subjects you suggested, thanks a lot!
  • rexo12
    Offline / Send Message
    rexo12 vertex
    Personally I'd like to see more detail on how you made your material maps, and you mentioned using substance in your tutorial, maybe you could talk about that. I'd also like to see some tutorials using free alternatives, you used 3D coat and Zbrush in the tutorial, maybe you can demonstrate the same process in Blender/Maya/whatever. That's probably asking too much though.

    Love your work, looking forward to seeing more.
  • Sebvhe
    Offline / Send Message
    Sebvhe polycounter lvl 4
    @rexo12 : I have to admit I'm not very familiar with Substance, I know it can be used for photogrammetry, I have seen quite a few articles about it. I'd rather link to these than have to change my whole workflow which already works just in order to use Substance. Again, I'm not saying Substance would be better or worse than Photoshop or any other software, there are many ways to achieve one result, Substance is just not one I'm familiar with.
    Which brings me to your second point, I do not personally think the software I use actually matter that much. I mean of course it's easier if you can follow precisely the same steps than my tutorial, but the whole process could be replicated with almost any alternatives. In the end it doesn't really matter if you use Agisoft, Reality Capture or some free photogrammetry software, what I do in Maya you could do in Blender, What I do in Photoshop you could do in Gimp...
    Do you think it would be too hard to translate the process to another software than the one I use?
  • janoshx
    Offline / Send Message
    janoshx vertex
    I would like to learn how you clean and make tileable texture with as many channels as possible.
    For example something simple like a tree trunk. It is easy to capture. After it is exported from agis photoscan I Import it to maya and make a simple cylinder close to the trunk. Now it gets interesting... I can use maya transfer map tool to project diffuse to cylinders UV easily but i have't yet managed to project UV or Height with satisfying results. I have tried this only few times now so there is still much to try. Of course if you want to project pine trunk the surface has very steep deformations(even negative) so in those cases the normal map would not work well anyway.

    To make long story short. Please make some tutorial how to extract more useful data from the scan. I have started to learn substance designer 6 and there is some tools (which i haven't yet much touched).

    Ah. If you have a trunk and you want to tile it you want to tile it only vertically. I think you can do this in substance painter with stamping it either on tiled plane or vertically tiled cylinder.

    aand... how do you delight the texture. I read that you can do it in unreal but how i could do it in maya for example.

    I can give you guys a some tip as well about the camera.
    Cellphone camera is okayish. You can me photogrammetry work if you have good camera in cellphone and you can manually control it with Camera VF-5 for example. Capture as DNG if possible, If not then PNG. The again I would not recommend to use cellphone camera anyway.
     
    DLSR is of course the best option. I prefer Canon 5Dmk3 with 24-70mm or 24-105mm lens. If you use zoom lens you must not change the focal length, ISO or aperture while taking the pictures.  I would recommend to use 24mm focal length and use duct tape to lock it in the beginning of shooting process. Use F8 and max ISO 400. Result wont go horrible on even higher ISO ( at least on 5dmk3). About aperture, don't go over F11. IF you use smaller aperture than F11 eg. F16, F24 or even 32. There is a phenomenon called diffraction involved in digital photography. If you use smaller apertures (bigger aperture number) than F11 you will get blurry image details because of diffraction.
    Good reading: http://www.cambridgeincolour.com/tutorials/diffraction-photography.htm

    You will get better results if you use monopod or tripod. Tripod recommended. Also it is recommended to use hand held camera trigger so you wont bump you camera while pressing the shutter.

    Fuji X100 and its newer versions are REALLY good photogrammetry cameras. It is an excellent camera for anything. Good 35mm lens, superb image quality and all the settings you ever need.  You can get second hand Fuji X100 with very reasonable price (around 500€). If you get x100 you have to get the latest firmware update. Originally it had very slow autofocus but the firmware update fixes that 100%.
    There is few x100 models like x100s and x100t at least. Get the x100, the others don't have anything so special for the extra price. They have a bit better image quality but nothing that would really matter. x100s and x100t are way more expensive.
    x100 is very small compared to DSLR and goes nicely in leather case in the bottom of your backpack.
    https://www.youtube.com/watch?v=FsDiFDBgiZw

  • Sebvhe
    Offline / Send Message
    Sebvhe polycounter lvl 4
    @janoshx : Most of your questions are already answered in my previous tutorial, altough I'm definitely going to add some new techniques. I wouldn't use Maya to bake maps and I wouldn't project on a simple cylinder, this will add to many discrepancies between high and low poly.
    But you'll find more information on my tutorial here (the free one on the right)

    However if you still have question after reading it, please let me know, that is what I'm trying to improve for my next tutorial update.

    Regarding what you said about camera settings, I couldn't agree more. I didn't talk about diffraction because you'll almost never have to go higher than F-11, but it's probably good to mention.
    As for ISO it really depends on the camera's sensor, mine is doing really poor, so I would never use ISO400, but it's just my camera
Sign In or Register to comment.