32 Comments

  1. Roland Rick Photography
    June 3, 2019 @ 7:06 pm

    To reply to your question: Shoot raw images, tweak them as good as possible (flatten contrast, rise shadows, drop highlights, get correct black and white point, apply s-curve to get back contrast ➡️ but now free of possible blown highlights, push clarity), export to a format the software can understand, and you get quite possible way better results. Check the 360° panoramas on my Facebook page (rolandrickphotography). They are all made of the 26 original DNGs the Mavic 2 Pro takes for 360° panos – hunt: you must activate “keep originals” first, otherwise you’ll only get the finalised jpg made by drone, which is way worse as the manually made one: high res 26000×13000 (manually) has more details than 8192×4096 (drone). Camera settings for exposure and white balance must be M(annually), expose correct to the highlights, this avoids missing details like the strings in the mills wings.

    Reply

  2. Matthew Bouton
    June 3, 2019 @ 7:06 pm

    Is it not better to video capture rather than take pictures.

    Once you have your video file. You could convert it into frames.
    The theory is that it would be of higher detail/ quality.

    Reply

  3. P Huster
    June 3, 2019 @ 7:06 pm

    I find the hundred photos shot straight on was a good start, but you really need 3x-4x more. Start at ground level, about 45* up, and repeat the process. This takes care of things left in shadow a straight shot doesn't get. Do the same thing around with 45* down angle. Lastly, the blades. You could capture one and apply the geometry to all 4. You'll miss structural damage, but it's a huge time saver. You need photos looking at the blades at the angle they are at, maybe 15* off center for a guess. Shoot straight at them, then 45* up and 45* down. Also, finding the owners and museums to sell your scans to is a quick way to pay for your hobby. I'd suggest saving to buy a 3D printer to make models of what you scan. Also, if you pay a bit more attention to the height, every time you change elevation, and attempt to take the 45* up/down shots at that angle, the mesh process is reduced. Less calculations needed. Lastly, the 45* changes to camera position must be done manually, move the lens, not the view. Optical zoom does a lot more for you than digital. Ignore the minimum and maximum distance though, concentrate it at 20% less than they state for ideal range. Good luck.

    Reply

  4. tommyb7973
    June 3, 2019 @ 7:06 pm

    Reply

  5. Илья Печников
    June 3, 2019 @ 7:06 pm

    why not just record the video instead of manually taking pictures

    Reply

  6. jose mañan
    June 3, 2019 @ 7:06 pm

    i did that once, i scan a arch-model with a digital camera. i had the same reflective error. tried photoshop in every image. but it was like forever.

    Reply

  7. Jan Kadeřábek
    June 3, 2019 @ 7:06 pm

    So you make a tutorial on something that is completely unusable in the end? Hmmm. Of course you show the workflow, but the info is too basic for me…
    Anyway, this example shows the main disadvantage of the photogrammetry – at this time it just cannot deal with any transparent or glossy materials or (as at this example) objects like fences / grass / etc. On the other hand, it is great for statues / rocks and similar solid matte objects that would be very complex to model & texture. It would be more effective to model this simple mill manually than make a 3D scan and trying to correct all those imperfections.
    These are my testing models (taken by regular camera): https://sketchfab.com/jendabek

    Reply

  8. Kyle Saunders
    June 3, 2019 @ 7:06 pm

    What do you guys think of Blender VR? Can you imagine such a thing?

    Reply

  9. Patrice Carbonneau
    June 3, 2019 @ 7:06 pm

    I do drone photogrammetry as part of my job as a University prof. Check this out: using drones to study rivers. Illustrated with a Blender video.
    https://youtu.be/U41WcufrCx0

    Reply

  10. Michal Burger
    June 3, 2019 @ 7:06 pm

    This reminded me of a really cool windmill model that I've seen recently, also made in RealityCapture: https://sketchfab.com/models/5a383d21a68f46859b93d3042f7627f9 . Interestingly this one was shot completely from the ground, no drones used.

    Reply

  11. Michal Burger
    June 3, 2019 @ 7:06 pm

    The number one thing you can do to improve your results is simply take more pictures, I would aim for thousand or more. Especially the areas with holes or missing parts, you should take more close-ups of those or of anything with slightly more complicated geometry.

    Reply

  12. Blender Rookie
    June 3, 2019 @ 7:06 pm

    I did the same thing with a DSLR using 3DF Zephry. It's a payware but they have a free version that works as good as the pay version.

    As far as best practices, the big ones are as follows.
    1) Make sure every image in focus
    2) Set the f-stop as narrow as possible for the lighting in order to have the widest depth of focus
    3) Make sure each and every feature of the object being scanned is captured in at least 3 images from 3 different angles. That of course can lead to hundreds or even thousands of images.
    4) Don't be worried about taking too many images. Good photogrammetry software will automatically disregard images that are too similar or will not help.

    Reply

  13. Steve Nilsen
    June 3, 2019 @ 7:06 pm

    Nice video going trough the process of photographing and using Reality Capture, but what about the end result? Is the final animation you did made using blender? What about a little tutorial on this? =)

    Reply

  14. Bathini Nikhil
    June 3, 2019 @ 7:06 pm

    pls make a tutorial about motion capture

    Reply

  15. jurandfantom
    June 3, 2019 @ 7:06 pm

    RAW photos, mask your object before CR software usage. If you use blender, then you can create displacement map from texture (alpha) to enhance details on your model just like that.

    Reply

  16. Nathan Reynolds
    June 3, 2019 @ 7:06 pm

    For taking the drone photos you can use drone deploy https://www.dronedeploy.com/ it is very expensive but there is a free trial.

    Reply

  17. Bruno Silva
    June 3, 2019 @ 7:06 pm

    Try addon OrtogOnBlender to import images.

    Reply

  18. The Internet Made Me Like This
    June 3, 2019 @ 7:06 pm

    Not good really :/

    Reply

  19. Danial Carroll
    June 3, 2019 @ 7:06 pm

    If you can afford a drone, I don’t think the cost of photogrammetry software will be an issue.

    Reply

  20. Blender Builder
    June 3, 2019 @ 7:06 pm

    Still waiting for part 13 of classic female character modeling 🙁

    Reply

  21. masterxeon1001
    June 3, 2019 @ 7:06 pm

    i was just doing this yesterday. I even used a DJI haha. Nice video! Reality Capture as well ftw.

    Reply

  22. Haze H
    June 3, 2019 @ 7:06 pm

    a few quick tips for photoscanning:

    – always take more photos than you think you need. (especially using reality capture, it can handle thousands)
    – if you want accurate(ish) albedos, get at the very least a gray card to get correct white balance, or at best some color checkers and create color profiles for each lighting condition.
    – if you have a choice of lens choose prime over zoom.
    – know the sweet spot of your lens (generally somewhere around f8 for most commonly used lenses between 24 and 50mm)
    – avoid fish eye lens
    – lighting conditions are everything, extremely cloudy days are perfect (some professional setups shoot at night with ring flashes for perfect conditions)
    – a good way to be more efficient if you can't shoot thousands of photos is to make a large shape pass, ie. get large scale photos without worrying about the detail level just to rebuild the shape of the object, then do a detail pass on a small section of the thing you are scanning, build a couple detail textures form the detail pass and use those textures blended together with a hand painted or procedural mask to texture your larger model. obviously this only works for uniform objects/grounds/walls/etc but it can be a huge time a computer resource saver.
    – lastly if you are scanning a surface (wall, ground, etc) don't bother reconstructing the 3D object and then baking down textures, it is an incredible waste of time. Use the orthophoto projection workflow instead and then use substance designer's scan processing nodes to extract your normal map and AO from the 32bit float height map (I believe you can also use xnormal for this but designer will give you a lot more control over delighting the albedo and generating your roughness map; but for a quick and dirty roughness map for those who don't have designer, make a curvature map from your normal map, invert a grayscale version of your albedo and use those as a starting point)).

    for a ton more info on scanning check out unity's scanning guide, it is extremely in depth and even covers specular capture.

    Reply

  23. Tom Kayak
    June 3, 2019 @ 7:06 pm

    And GPU is essential! 8 thread i7 … 4 hour's, gtx 970… 17 minutes 😮

    Reply

  24. GalacticSalmon
    June 3, 2019 @ 7:06 pm

    I've done a few 3D scans, some with just my phone and some with a Mavic Pro.
    Not idea if the Phantom 3 has the exact same software features as the Mavic has, but when scaning an object such as the windmill the easiest thing would be to use the "Orbit mode" to automatically orbit your subject and just use the camera timer feature to take a photo every x seconds. Think of it like making a slow turntable animation. Then manually fly around and get some closeups of smaller details, such as the blades on the windmill.
    Using something like DroneDeploy can also make taking the pictures way easier as you just plan it's flight path, then it does all the flying and photo taking.

    For something like that windmill, if I wanted as high resolution result as possible, I would probably go with 36 photos in one circle around it. Then probably another 36 photos every 2-ish meters up and then fly in a grid over the place to get some top down photos of both the windmill and the ground around it.

    For anyone just wanting to test out photo scanning I would recommend just going out with your phone and take a ton of pictures of a decently sized rock or a tree stump. In the right light conditions that can produce super nice results.

    Reply

  25. Jim Cochran
    June 3, 2019 @ 7:06 pm

    I’ve been doing this for a year now with P3A and Photoscan Standard. Latest release will pull photos from video so you just fly it without lining up each shot. Works great.

    Reply

  26. Stuart Eliason
    June 3, 2019 @ 7:06 pm

    If you didn’t already in your search for free software, you should check out Gleb Alexandrov’s video about free photo scanning. It’s pretty good.
    Also, can your drone record video? If it can, could you have flown the drone around the windmill and then used all of the frames from the video to scan with? (Maybe the motion blur would have messed things up.) Cool tutorial and idea keep up the good work!! Any idea when the next CG Geek video will go up?
    https://m.youtube.com/watch?v=GEAbXYDzUjU&t=4s

    Reply

  27. gower1973
    June 3, 2019 @ 7:06 pm

    Looked a bit of a fail to me the geometry was a mess, You could of modelled it from a few reference photos in the time it took to go there do the drone stuff, buy the software, learn the software wait for it to compute the point cloud etc. The only advantage I can see is generating a automatic texture map

    Reply

  28. Sir Brokoli
    June 3, 2019 @ 7:06 pm

    If you want to take the free path Gleb Alexandrov has a great tutorial on it 😀

    Reply

  29. Ganapathy M
    June 3, 2019 @ 7:06 pm

    Why are you using chap stick

    Reply

  30. Matthew Strong
    June 3, 2019 @ 7:06 pm

    Try scanning like a rock or log to improve your skills. I have a YouTube playlist on this topic (photogrammetry) since I've been curious about it as I have been learning about everything related to 3D software and stuff over the years. Also Agisoft Photo Scan is a great alternative to Reality Capture and might give slightly different results which would be helpful to maybe compare so that people can decide what they should buy in regards to professional software. (P. S., your video is now on the playlist since it will likely benefit from joining it's friends on this topic.)

    Reply

  31. Alkinoos Androulakis
    June 3, 2019 @ 7:06 pm

    Great video, Great presentation Cool tech. Keep up the good work !!!

    Reply

  32. nosferatu5
    June 3, 2019 @ 7:06 pm

    I dig the camera mans mute sense of humour.

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *