My grandfather was a photogrammetrist back in the 60's until he retired in the late 80's, He died when I was young but it's amazing to think that he spent half his life working with the same mathematic formulas they use for measuring photos to print models in this video lol
Bookmarks for me. 0:30 Recap 02:10 Taking Photos 04:15 Standard Reconsturction 04:50 Hit Start Or Structur for motion out Tip 05:30 Troubleshoot capture issues 05:45 when Happy Complete Cpmputation 05:55 Notes Turn Green as they c0mplete May take a while 06:12 Double click TEXTURE BUTTON 06:30 Export Model to Edit 06:45 Augmented Reconstruction 08:14 Live Restruction 09:10 All Mesh are Hollow-Plane Cut 09:30 Change Scale 09:43 Sculpting 10:03 Wiki Wiki Time! step 1 Drag photos into Programme Step 2 Right Click Structure from motion-Compute Step 3 Wait until computed Step 4 When Happy Finish Compution Step 5 Double Click Texturing Note to View the model Step 6 Export Model Step 7 Plane Cut in 3D sotware Step 8 Change Scale to suit
You prolly dont care at all but does anybody know of a trick to get back into an instagram account? I stupidly forgot the login password. I appreciate any assistance you can give me.
I did this with a local cathedral (with permission, of course). My results then weren't great however, so I may have to try reprocessing the images with this software.
There is a Litchi software for DJI drones that can do that. You just set the waypoints and camera angle and it should work (but be careful if you use that)
@@logitech4873 you either need a better camera or fly closer. If you used Photoscan or even RealityCapture, you will not see improvements in other software.
This is the most amazing thing I've seen in at least 24 days. Now I don't know if I want to 3d print something or put it in Unity... or whether I want to mod my house into GTA5... or build a rig that automatically rotates the camera around an object.
@@talharoxen you mean putting the 3d model into GTA? You use the mod tools, I forget which one you use to export to rockstar format but it should be easy to find. There's 5 or so tools in total that help you mod, it's not as easy as a Bethesda game but it's not that hard thanks to lots of documentation and community support.
Note: Always take note of what focal length you're shooting at - take not of what lens, focal length, camera type and settings you're using. If your scan comes out wrong it can help you when having to enter metadata manually if a problem occurs. It's especially important to make a note if you're using a smart phone camera because often vital data such as focal length is not recorded in the metadata - which is some cases makes the pictures unusable if you can't figure out the focal length and enter it manually. I can strongly recommend people using the following trick: Import the photos into a program like Photoshop or GIMP and add a mask around your object removing detail in the background. You can also zoom in if you couldn't get close enough to the object in the first place. Remember to retain all metadata while exporting your photos.
you might be interested in GL Intercept github.com/dtrebilco/glintercept here is one possible application (note this guide is quite outdated): brian-horning.com/google-earth-to-obj-using-glintercept/
Awesome, i know exactly where that first statue is, one of my favourite spots in Prague. I go to the cafe in the National gallery and that restaurant to the right all the time :)
If it's a small object you could use armature wire to hold it up so that you could see under it. In any case, try to put the model on a base that is smaller than the model.
He touches on it in the video, but the single most important thing you can do to achieve high-quality results is to use a good camera. Yes, you can get good results with a Gopro, cellphone or point and shoot cameras, but if you compare a model generated from a GoPro to a model from a DSLR/mirrorless camera using a good prime lens, the results from DSLR are absolutely superior. When comparing models between lower-end cameras and high-end cameras, look at sharp corners/edges, photogrammetric reconstruction with lower-end cameras struggle to produce sharp corners, they come out more round. If possible, use a full-frame camera. Make sure you move between shots, no standing and pivoting. Take steps in between photos. Take the photos in a systematic way: go low, medium, and high all the way around. For tall structures, you can also scan vertically (take photos as you move up the structure, scoot over a bit then come back down until the whole thing is covered). You want to disable autofocus as it causes focus changes between photos and will cause pixel blur when the algorithm looks at different photos of the same feature. Set focus to infinity and leave it there. You want to achieve high overlap between successive pictures, both vertically and horizontally (like 70% to 80% overlap). More overlap means more matched features means a more dense point cloud which means a tighter mesh. If you want to know how good a job the software is doing with your photos, turn the textures off and look at the raw surface. The texture hides lots of artifacts. Use a fast shutter speed as well to take motion blur out of shots if your shooting by hand and not with a tripod (this is especially important if shooting from a drone). Use a polarizing filter to help reduce glare if shooting outdoors. Like hard shadows, the software struggles with how to handle it. Post-process your photos to increase contrast in shadowed areas.
It has advantages, but also disadvantages. This is why we combine different scanning techniques for perfect results. Great video and great progress in the software!
@@mageenderman Well, I dont work with photogrammetry, but I have read that they also use laser scanning. Its the most precise way to scan a object and the results are outstanding. Here is a video by Konami scanning Schalke 04 stadium and showing previews from the game: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-qAkUF4Kbgxk.html
Just when my modelling abilities were starting to improve, this comes along and makes modelling obsolete. I've wasted my life. It is rather cool though. Can't wait to try it!
Hey they're not obsolete! For mechanical parts, CAD is still the best option. And for organic shapes, you'll be really happy to know basic modeling when working with photogrammetry meshes, they almost always need a bit of editing/clean up before printing :)
@@Prusa3D It's good to know I'm not worthless just yet lol. It's funny you should mention CAD, as I recently started using CAD to design machines, and have also sucessfully made game assets from CAD models. All of the metal targets in my FPS game are actually real-world products that I produce. The game models are just slightly reworked CAD models.
Awesome video! Would be nice if you do a live tutorial on 3d reconstruction of more problematic objects, like ones with shiny and flat surfaces without texture. I.e. like reverse engineering old injection molded parts with glossy surface and flat color, or something like this.
Oh wow, I've recognized the accent and the statue of Dvorak, then I found the video about your printing farm. Great job you're doing there! I wish I could visit the farm one day and film interview with you!
Yes meshroom is great! I tried to use meshroom at work to create a reference model of a complex assembly, but management was too stupid to understand why it was necessary. Let's just say the original documentation was useless for building something with thousands of parts. A 3D model of the entire assembly could've been built with photogrammetry in a couple of days, then distributed to every assembler for reference. NASA did something similar for their space craft where they always built two of everything. Photogrammetry is a much more affordable alternative to this.
Thank you for the video. I was new to the topic and tried to read up on it, but your video was just perfect to get into it. Especially the hint that you can make ideal photos on cloudy days has helped - on cloudy days it is not lacking here with me anyway...
The live reconstruction might be fun if you have some sort of file syncing tool. Take pictures on your phone and have the model already processing by the time you get home.
Great video!! Thanks for introducing Meshroom to us and a tutorial on how to use it for photogrammetry :) I was going to start doing this soon with Autodesk ReCap, but this might be a good alternative as well.
There's still plenty of room for improvement, especially when it comes to speed and ease of use. But yes, it's pretty incredible how fast did the technology improve :) - Mikolas
Is there any way to use this technology with an older mac? i have a MacBook Pro (13-inch, 2017, Two Thunderbolt 3 ports) All Photogrammetry programms ive tried wont work with my system. Very grateful for any tips, or programms that i could use
I've tried Meshroom because I wanted to do a tutorial on it but I found it complicated and really had trouble getting good results. These days I use Reality Capture. It's not free but pretty cheap.
@@dudeabides23 - Indeed. I learnt photogrammetry from a guy called Sid Ray, who I suspect had a fair bit to do with military photographic interpretation around that time.
Very cool. thank you for letting us know about this program. I've not been super happy with photogrammetry since 123d Catch was taken down and this looks like it may get me back in to it again!
I've used PG on old footage from movies or documentaries... Comparing the models made from old footage with models made of the same scene with newer footage reveals subtle changes in shape. Besides buildings and construction, it is fun to see a hillside 'heave' during the rainy season and 'shrink' by the end of summer. Sure, it's only a few feet out of thousands of cubic yards, but is can be measured! Similar things are seen with Grand Canyon footage; old fly overs and newer ones show lots on interesting 'shifts' in form. Are volcanologists using this?
The traditional way of 3D modeling real-world objects is time consuming, so an automated, easy and affordable way of using photogrammetry that reduces production times where 3D modeled objects are used is a good thing.
Wow, uprime videl sem thumb asi pred 3 tydnama nebo kdy a az ted sem se k tomu zase nahodne dostal a koukam ono to ma pres mega sheldnuti a je to cech :D ... no nic, tohle pokud se mi vysledek podari dostat az do UE4 tak to bude super. Diky moc za info ohledne meshroom!
I love doing photogrammetry. The only thing that I wish was different is that there were more software that runs natively on a Mac. I know Macs are a smaller part of the overall market. Before anyone says "Get a Windows machine!" I'm planning on adding an Nvidia card to my Mac Pro and make my hard drive dual boot to have both macOS and Windows. I just know Mac better than Windows right now.
For those that dont have Nvidia & don't have to means to get a graphics card. you can use a virtual machine in Microsoft Azure that has it. it comes about 60cents per hour
I have an idea, with the same setup with the white background, just use a motorized rotating plate in the bottom, and use an automatic shoot every 1 sec
What if there was a add-on camera the attached to print head to reconstruct an object on the print bed? You already have a machine that can move across the axis. You place an object on the bed choose scan the printer takes as many pictures as needed then you choose size and print a copy. Next step x-ray.
Imagine how far we will have come in 20 years. Conveniently scanning everything within seconds with our "phones", if they are still called like that in the future.
Presumably it uses background details to work out where photos were taken from and that's why rotating the object on a white background doesn't work well. So perhaps it would work if you place the object on something like a chessboard on a turntable. That way you get a background it can use as a reference.
The further & closer the objects in the scene are the better. Having some buildings or trees far away is great for matching, even if they're not what you want to reconstruct.
Somedody might suggest a blog, forum or fb channel where to talk about meshroom, and ask suggestions? I tried some attempts but i can't get a mesh done, and I don't know where is the problem:( thanks! This is a very well done video and thanks :)
Some advice on photogrammetry: 1-Use softboxes to illuminate your scene...lights must be flat and casting the least hard shadows as possible. 2- use a solid background..white, black or green..ive gotten good results with blackand very little image retouch. 3- mask or use chroma key on your pictures..this will eliminate all the innecesary information from the image, letting software create tie points only from the points of interest. 4- use a DSLR..really. Most DSLR equipments include important exif data such as focal length, iso, shutter speed and the type of lens used...some of them can even estimate yaw, pitch and roll..which helps most photogrammetry softwares compensate errors in the alignment procedure by using such camera info...effectively tying more points from the pictures overall. 5- use a turntable...keeping your camera fixed in one place and rotating the object of interest can produce amazing results..but due take in account you must forcibly mask your picture...this process can take some time...take in account that there is a point in which taking more pictures will not make a significant difference on the fidelity of your model..you dont need to go up to the fivehundreds or thousands of pictures for something you can easily reconstruct with only 30 to 50 pics. Keep it simple. 6- Postprocess your photos...i recomend using adobe bridge with the camera raw add-on to increase the contrast of multiple pictures simultaneously..this will help software distinguish borders more easily and an overall result in a less noisy point cloud. Just remember to not clip your highlights or crush your shadows too much. 7- Reconstruct in a low or medium quality so you can save time...if you are conviced with the results..simply repeat the process in a higher setting..this will help you save extra time..since medium reconstructions can take only minutes, you can realize faster if the photos you shot were good enough to really reconstruct your model or if they were a complete miss.. 8- Play with your settings...overall experiment...photogrammetry depends on multiple factors...photo rig..lighting..the complexity and material of the model..background..image noise..photography postprocessing...keep tweaking until you amount enough experience to realize how to have smoother workflows. 9- You will make mistakes, learn from them...my first reconstructions looked like chronenberg monsters, very little tie point clouds, bad lighting, bad photo quality...you will learn from these mistakes along the way. Today I can reconstruct models up to a million+ poly count with realistic textures and so can you. The important lesson here is that you learn a long the way and keep trying new things.
Unfortunately, we can't agree with some of those. Especially for Meshroom! 2,3) Do not use background and definitely do not chroma key your pictures, this decreases the quality significantly. We talked about it with the developers: "the results are worse because we will only have points on the object which represent a small sub-part of the image, so the camera parameters are less constrainted which leads to less precise results" 4) EXIF data are included in pictures from phone for almost a decade, we got excellent results just by using smartphone camera 5) No turntable with Meshroom
Noob question!!! How can i put a design on a constructed image? I would like to get the 3d design of a motorcycle and play around with colors etc to see how it looks like
