Which photogrammetry tool is the best ? (3DF Zephyr, Metashape, Reality Capture, Meshroom)
Vložit
- čas přidán 13. 05. 2024
- In this video I capture more than 600 photos of an old barn/shed with the DJI mini 3 pro and convert them to a 3D model using photogrammetry softwares. But which one yields the best result ? Which one is the fastest ? The easiest to use ? The cheapest ?
I propose a detailed comparison of four of the most well known photogrammetry softwares on the market : 3DF Zephyr, Metashape, Reality Capture, and Meshroom.
I used the default settings for all the softwares in order to get a fair "I'm new to this" comparison, where someone new to the field wants to try it out without much prior knowledge.
00:00 Intro
01:07 Capturing process
02:52 Nice interlude, dude
03:15 Results comparison
03:37 3Df Zephyr
07:06 Metashape
09:34 Reality Capture
12:46 Meshroom
15:49 Side-by-side comparison and discussion
20:50 Pricing options
24:25 Side-by-side comparison with groovy music :)
Music credits :
Beautiful Nightmares - Adelyn Paik
Boo (Instrumental Version) - Adelyn Paik
My Kind of 2 A.M. (Instrumental Version) - Adelyn Paik
Leopard - Andre Aguado
Let It Breathe (Instrumental Version) - Adelyn Paik
Sleepy Hungry - baegel
Sleepy Tiger - Adelyn Paik
The Moonlight Serenade (Instrumental Version) - Adelyn Paik
Twilight Sky (Instrumental Version) - Adelyn Paik
![[Vowel]물고기는 물에서 살아야 해🐟🤣Fish have to live in the water #funny](http://i.ytimg.com/vi/3G9MH063Cx4/mqdefault.jpg)
Hi. I'm only just becoming familiar with photogrammetry, so I REALLY appreciate the time you took to produce this video review comparison. It was extremely enlightening.
Regards Mark in the UK
A beautifully produced video. I’m keen to see future more in-depth videos on how to use the individual programs
This is exactly the video I needed. I'm wanting to try photogrammetry, and this is very helpful.
Phantastic Review! Thank you for all your time and effort. Cheers Martin
It's best to show meshes without texture. Every software can do texturing so the mesh difference is what matters.
Thank you for the introduction to photogrammetry. Now I have a basic understanding thanks to your efforts. Nicely done sir!
Thanks for such nice and fun to watch comparison :), cheers!
Very good explain and comparison ❤
Thank you! Great review and well done!
Hi there, thanks for the comparison. The reason RC made a single 8k texture is because that is the default. To get the best texture, make a new unwrap using the unwrap tool using "fixed texel size" style with the texel size set to "optimal". Then texture again.
Thanks for the tip !
very good comparsion, thanks for making this high quality video!
Great video. As I was waiting, I thought that it it would be really cool if, next time you compare software or other things, you could have each software's stats (time to generate the model in this case) over the video of the respective softwares instead of as columns from left to right. It would also allow you to update the data on screen making watchers more engaged.
Most programs use the same algorithms, so you are more or less testing the default settings. But for noobs this is valuable information. I used Metascan until 2022, and then I changed to Zephyr. Results are pretty much the same. It's the input that defines the output via the settings (metashape has no holes in the roof, but fuses the planks, zephyr has hole in the roof but seperates the planks) . Drone shots are always challenging. Good sharpness and depth of focus are paramount. But so is lack of noise, with a small sensor camera that is always challenging. Good work, BTW.
curious why you changed from MS to 3DF-Z
Thank you for that laborious comparsion! I think it is difficult to compare them with "default" values. Some software closes, lets say, 10% holes of the mesh, while the other app clothes 80% of it at default. So, maybe you append a "part two" here with the best presets you found out.
For vertical infrastructure, bentley's Context Capture was amazing. I think they bought it from acute a few years back. Might be screwing up the names, been forever since I've used it but I was the most impressed with it for anything going vertical. But great comparisons. I'm only familiar with Meta and Reality Capture, Meshroom scared me when I started playing with it.
Thank you for making this video, it is very informative. Once you come to the comparison of the four apps side-by-side, your focus is on the features. That many images aligned, so many triangles produced... But more important is how do you overall like the end result? It depends on the application of the model of course, but taking everything into account which package would you go with to model existing buildings, with a good, detailed, small-scale texture, to allow lighting to be added once the model has been made and imported in 3D design software? Thanks.
Great video for the newbies in the photogrammetry world! Thanks for the thoroughly comparison and keep the good job. i’m a civil engineer and usually work with lidar scans for field surveys of buildings. Can you tell if all the models that you make could be usable for measurements of element dimention and distances?. The lack of detail in the barn truss in meshroom model seems to be a negative point. Do you believe that using the doble of resolution could bring this elements into the model?. Thanks again for the great video 🎉.
Thank you for exclusive quality video
Thank you ❤ pls try tutorial for 3df Zephyr ❤❤
Great comparison.
Thank You ! 🔥
Great Work! I really like Metashape, and its predecessor, PhotoScan, I´v been using since back than. I think is very fast even for testing and there are some "hidden" features, as map transfer based on the original photos used (pretty cool).
Now, seeing all the other options i can be sure that the Metashape fits better to me. Thank you
Nice comparaison 👌
Voici la vidéo parfaite que je cherchais pour me décider vers quel programme me diriger. Un grand merci. Salutations de Suisse. Christophe
wonderful introduction！ Thanks
Automatic like! I've been teaching myself this skill through YT university and your video hit the spot.
I purchased Agisoft Photoscan in 2014.
I don't pay yearly or monthly subs.
I get updates ALL the time.
The software KEEPS getting better and better.
Best 179$ I ever spent.
I own it, I can reshoot objects and add the new shots to 're-do' my final 3D models.
All for a base 179$ USD. All the upgrades and new stuff is there for me.
No 'credits or extra charges'.
This is all my opinion. However, I spent 179$ and I have 1,000s of models today.
To me, that makes my work more 'original' than most.
You will not be able to 'spot' a 'generic mega-scan ' model in my work.
That is because I can scan whatever I want and 'keep it real'.
@KAP01able How do you scan with this software (agisoft)? Do you have a list of camera or hardware to use it? (please) (from cheapest to more expensive)
Thank you so much
Ok, the photos --> crappy high-poly mesh is an interesting step, but much more important is the crappy mesh --> usable mesh + texture. Can you do a review of level of automation and quality of this step for diff packages?
You should compare the point cloud not the mesh, since mesh can be done with many different parameters from the original point cloud
I agree with you it's a very interesting job, but with reality capture you can get much better results.
Just a tip for the cinematic shots at the beginnig. Buy a ND Filter Set for the Mavic so you can use the 180° rule for your shutter speed. The shots look really "jittered" because you used a really high shutterspeed.
You guys don't understand what photogrammetry is... It's miles away from what you guys do with video
@@trollenz i meant the cinematic shots in the beginning, not the photogrammetry stuff. I know what photogrammetry is and i know that you don't need a nd filter for it.
Usually for serious tasks where I need quality, I choose Metashape. It collects sharp, accurate meshes. Sometimes I try to work in Reality Capture, but it often cannot glue an object into one part and splits it into several components, and gluing them together manually with dots is much more time-consuming.
When comparing these two programs, I understand that Metashape is more geodetic software and of course the quality is more serious. Reality Capture began to lose in the last two years, although in some moments it is more convenient. But convenience does not equal quality. Despite the fact that I work in Unreal Engine and export my meshes to this engine, and of course RC exports it to the engine more conveniently, while in MS I have to tinker with dimensions and pivots. In general, choose what is convenient for you for a specific task.
Cool video, but for the perfect comparison there would be great to take the shots during the overcast.
Thanks a lot!
worthy! your system specs would be of interest, too. Ram, Proc, Cuda cores for example.
sud ouest de la France ce type de grange ! merci pour la vidéo
Hi, really interesting results. I didn't know until today Meshroom existed. Could you do a test like this with OpenDroneMap as well?
I loved the comparison, very complete. I just want to comment that I just saw that Reality Capture will ditch the PPI model and offer free versions for students/hobbyists and $1250 / year for companies making >$1M per year
Où est-ce exactement ?
Merci
One important thing to understand… never ever use metashape with default preset buttons. Try to always use custom numbers for poly and tie points. Never generate the dense point cloud. I usually run 2- 5 million tie points on camera align step. Then you need to use the gradual selection masking to delete failed points in all of the categories like reconstruction uncertainty.. later generate the mesh only by using depth maps.. set the mesh density to custom and at least 10 million polys.. you can decimate it later. Have a good scan
Jerome, thank you for the detailed, objective comparison. One question: Can you clarify which *versions* of the software were used? For example, 3DF v7+ is completely different than prior versions, making this a very important detail. Thank you!
i think the comparison would be more precise if u had single rendern engine for all 4 models, furthemore there are a lot of settings that can be tweaked during the process of aligning photos and camera positions, in all programs that can misslead the final result
great video tho)
can you make a video tutorial for the shooting process only?
ty for the comparison.
27 4k (216 mpix) textures is only "double" from 4 x 8k textures.. yes.. they are so many either way
Good video!
Good video. Thank you for taking the time to do this. You mentioned the excessive time it took Meshroom to process this. I'm very curious, what hardware are you using to get these times? CPU, memory, graphics card(s). Knowing that would be valuable to the viewer.
Thanks again :)
I think he is using an RTX 3060 GPU.
(He had this card in one of his other videos.)
I'm using a RTX 3060 GPU, as lucu lucu mentionned, along with a Ryzen 5800X with 32 GB of RAM. All the files were hosted on a 2.5" SATA SSD Samsung drive. The same machine was used for all the softwares with no other background task running so it should be relatively fair of a comparison.
@@jerometabeaud9400 Thanks for the reply. You have a decent setup. Meshroom must really require some horsepower. Odd that the other software wasn't as hindered considering it's doing pretty much the same thing.
Hi, thank you for the video. I have one question. I watched several videos about drone mapping. and most of them used RTK models which are too expensive start. but when I watch this video I think you can make a good 3d map or models even with mini 3? in my case I want to try single building like this video or 100m x 100m open field in national park(i do have permit). Do you think mini3 pro can make it?
Drone mapping and 3d modeling are different things. Generally in mapping, the accuracy of the end result matters the most. You should be able to measure distances on the final map with very high accuracy. That's why RTK drones are used.
Technical info: GPS singal has some inanncuracies, especailly when measuring heights. An RTK station measures the GPS signal at the same time as your drone flies. Since the station is stationary, you can "fix" all the measurements to be stable. Then you apply all those corrections to the drone's GPS position. Therefore you get much more accurate results.
great comparison but could you please mention the versions of all softwares you are have used?
Fantastic, informative video, mate. THANK YOU! I'm a starving artist, so I'll stick with Meshroom for now!
Very interesting video.
Would it also be possible to compare PhotoCatch, which uses the Apple API ( Apple Object Capture)?
I heard it produces quite decent results with much less processing time.
Also, would you mind if you share the data set? Would love to do my own photogrammetry of that barn.
Thank you !
I do not own any suitable Apple device (phone or iPad), so I won't be able to compare the results. From what I understand of this app, it makes use of the new lidar on the latest versions of the IPhone, in order to speed up the results. It mostly works in close proximity, so for a building like the one I scanned in the video, it might be a bit cumbersome.
For the dataset, I'll see if I can find it and post a link to it. I'd be interested in seeing what you can achieve by tweaking the parameters ! Keep in mind that I did not completely cover all the parts of the building, and some photos are grossly underexposed/bad.
Great video !! I have a question, I have done a render with Meshroom, and have all the maps with the mesh, etc... on my drive.
But when I load the mesh file on 3d Online Viewer it only shows me the grey version. I can't seem to understand how to export or upload my project online (for free preferably). Just curious if you use Sketfab or any other particular viewer like in this video?
You need to open the 3d file generated by the "texturing" step, which also stores the UV information to load the textures. I used Blender to render the result, but any 3d software will do fine as long as you feed it the right file.
Hi, thank you for this great comparison! Not directly related but to the process: instead of taking multiple single photos, how about taking a clip and splitting it into photos later. The resolution of each might be lower but the amount of information might be override this. I would be interested in hearing your thoughts
When the drone takes a photo, it embeds the geospatial info within the file: lat lon, camera data at time of image, etc. Stills extracted from video will lack the geo data required to be ingested by these photogrammetery apps.
@@02CanGT It isn't a requirement at all to have any image metadata, even camera intrinsics can be inferred from the images. Meshroom can take a video and use every n frames, it even has a motion blur filter that removes images that appear blurred.
While having lots of images is generally a good thing, it's going to struggle with the lower resolution of a video. Better to have fewer high res images than a load of dross.
Great vid
But what to use if your gear is MacOs?
This is an interesting video but it's impossible to make a choice based only on you using each application with the default settings. It would be much more useful to see a comparison after you have got used to the software and used more optimal settings. It would also be very useful to know your PC specifications, especially CPU and memory, to get an idea of approximately how long processing might take on a different system.
Would you consider redoing the processing with the same dataset? Thank you.
"would you mind spending 6 months to fine tune every piece of photogrammetric software available on the market on various projects and let us know the results" 🤣🤣👌🏻👏🏻
very good video
Do you know if it is possible to use Mushroom on a bunch of digital pictures... pictures without shadows or backgrounds? I have 60 images of a 3d rendering from every angle and I would like to convert it back into a 3d rendering. Thanks
Your pronounciation is very good for a french guy!
Have you tried comparing these 4 softwares with DroneDeploy and DJI Terra? Also, maybe a review of the DJI M3E since it is touted to have a survey
-grade output?
Really ought to upload in 60fps, alot of the panning shots are hard to follow due to the stutters.
5:00 and 6:00 example
i like this!
to make MetaShape less blurry >
- generated masks based on the actual mesh model,
- estimated image quality for all cameras (with ignore masked regions option enabled),
- disabled cameras with image quality lower than 0.8,
- disabled cameras looking at the model from behind,
- rebuilt texture.
A naive question concerning the preparation phase (taking the picture): did you do everything manually or are there scripts you can use?
I really like Meshroom (with a powerful computer). I also use WebODM in Docker (Linux). Also free and great.
Excellente vidéo, je fais la même chose que toi Jérôme, mais sans faire la publicité à Metashape. Moi c'est pour mon travail.👍
I just started the video and noticed that you were shooting at f1.7 which gives you a very shallow depth of field and makes it tough for the programs to recreate a sharp image. I would try f8 and raise your ISO so you keep at least 1/200 shutter speed. I think you will find great results.
Difficult to do this on a small sensor at reasonable ISO though
@@MakeTeachRepeat try it and see
@@MakeTeachRepeat what do you consider a "reasonable" ISO? because f/8, 1/200s, and 400 ISO yields an approximately equivalent exposure, and those seem like pretty reasonable parameters to me. and if the shutter speed starts to become a worry, there are plenty of f-stops between 1.7 and 8
Mini 3 Pro doesn't have a variable iris. You can't change it from 1.7. this it totally fine for a drone. Also, the camera is on a literal gimbal, so if you make slow moves you won't get any motion blur or shakes.
@@jotch_7627 The only reasonable ISO for any camera is its base ISO. This gives you the most dynamic range and the least noise. For the Mini 3 Pro it's 100 and 500.
What about pix 4D! Please could you try that as well
Hy, Jerome, can DSM (Digital Surface Model) by using stereo satellite image which has only two pair of image in it using meshroom like metashape and envi or erdas for that matter can provide, if yes can you provide me with a steps or resources to achieve my goal, since I don't really have much experience in Coding and all the open sources methods involves coding, meshroom looks promising but i have my doubt i don't want to waste my money ordering satellite imagery before conforming the process.
Where in the world ?
Would love for you to compare WebODM. Also free, not sure if it's open source.
If I have a video recorded of drone footage and I export all of the JPEG images from it, can I do a scan using those images?
You can, but it won't be easy. It will take A LOT of time since the software will have to position the "frame" somewhere in space.
You see, one good thing with photos taken from a drone, is that they have embedded their geolocation (position, height, angle of camera e.t.c.)
Also if the given video was shoot in a fast pace (and therefore blur is present on the frames) it will also make it much harder for the software to process.
@@djgeorgetsagkadopoulosyes and no, you don't use GPS tags for pair selection but since it's a video, sequential can be pretty fast as well
What program did you use for displaying the model and how did you export it? I can’t seem to be able to display it like this
probabily blender
Hi I am only one day into finding out about all this. (I'm also still busy trying to select the drone which is right for me).
If someone could answer this I would be very grateful. Q: What would be the best software for a beginner with high aspirations to achieving professional results?
Will this process work on a very large object, like a small mountain?
What hardware do you use? Especially which GPU. Which GPU would you recommend for photogrammetry?
What is your specs ?
Can you give me the original video of the drone and let me try another software for mac
Which version of Metashape did you use, Standard or Pro?
How many samples in Metashape?
Is thdere an alternative that doesn't require CUDA cores?
Just a comment as a professional photographer: the green on the more outside bits under of the model is because sunlight bouncing upwards off the grass is green. That’s more just something you learn as a photog after doing it for 20 years. Your own brain has auto white balance and you won’t notice it often.
You can easily combine the components in Reality capture using control points...
Connaitre les specs de ta machine serait utile pour interpréter les temps de rendus, bravo pour le reste!
I also do some photogrammetry and to be honest metashape is ok only the correct settings I have some lecture but they contradict themselves...
Hiw can I compare if you lower the quality of the last program how silly
How many usable pictures did you end up with?
OpenDroneMap! Best bang for the buck!
What about LumaAi?
what are your pc specs?
Might you be able to make a video about how to make videos of your 3d models? How do you get to move around your virtual environments so consistently and smoothly? I would like to demonstrate some of my own models to my employer and friends and a small well drafted video would do it quite well.
Godd idea, I might do a little tutorial about that in the future.
In a nutshell, I exported the 3d models from each software and loaded them into Blender. I then animated a camera and an invisible target, and then rendered the animation with each one of the models, separately. It took way too long for what I wanted to do, but it does look smoother and makes it easier to compare the different meshes.
What about WebODM?
I never tried it, I should give it a go someday !
Why not redo the Meshroom test instead of just putting it into the video with faulty parameters?
Reality Capture is not only backed by Epic Games and the Unreal Engine but will also soon be completely free to anyone with under $ 1 million in revenue.
So... learn Reality Capture. It will soon dominate the competition.
A point barely mentioned, GPU usage. Nvidia support doesn't help if your workhorse PC has an AMD GPU.
Reality capture just went fully free
what about nvidia nerf
Metashape is a forbidden 403 so that’s not installable anymore
Thank you for the testing, but the verbal conclusions had almost zero value. Reality Capture is the king of all as seen from the Video Captures. Spitting out numbers does not speak of quality, of which RC is the absolute beast. Dedicated talk on edge quality, geometry quality (straight edges when there) is the most important thing, as well as texture tone quality (the way the software outputs colour and shadows/highlights). Leaving out edge and surface quality is like talking only to people that will retopologize the entire model, which is 1% of the users, at best.
I think you didn't capture the texture on Metashape at all. The colors you see are just the vertex color. So the comparison is not relevant...
It just isnt very fair to do comparisons when just using default settings.
To my eyes Metashape seemed the most competent, but you didnt up the texture resolution to match zephyr and the others.
Likewize, zephyr generated far less triangles than Metashape did, so zephyr's mesh isnt truly representative.
you forgot the best one which is Itwin/Context Capture