is Photogrammetry the 'big thing' we should all be learning?

oraeles77

someone elsewhere on a different post suggested Reality Capture, it looks damn easy to use, however it looks helluva expensive.

and I've come across Colmap, which appears to be free/opensource.

in your opinions would you say Photogrammetry is an important skill to have in your CV?

Andreicus

Try out 3DF Zephyr, there are a lot of different versions for a reasonable price and it is perpetual.

Anyway it is better to have something more instead of not having it when it comes to CV in general. 

renderhjs

There is also Meshroom from alicevision with a node based interface, its also open source and available for free.
https://alicevision.github.io/
Review
https://scanbox.xyz/blog/alicevision-opensource-photogrammetry/

overview video
https://www.youtube.com/watch?v=v_O6tYKQEBA

There are no binaries up yet so for those that don't feel like experimenting with Visual studio you have to wait a bit. There is an active release discussion here:
https://github.com/alicevision/meshroom/issues/128

musashidan

oraeles77 said:

someone elsewhere on a different post suggested Reality Capture, it looks damn easy to use, however it looks helluva expensive.

It's $90 for 3 months. It's dirt cheap. It's not as if you need it activated at all times.

Photogrammetry has its place, but it's not essential learning by any means. Top substance designer artists are using scan results(especially height/normal data) as reference, to great effect. Recreate the real surface and still have procedural power to create many, many variations.

SebKaine

I agree, if you are in production environment, you can learn photogrammetry quite easily in a matter of days or weeks. 

I recommand using both reality capture and photoscan , each has his +/-.
I love Reality capture for organic stuff it gives the finest details.
I love Photoscan for hard surface mechanical object and for very tricky manual alignment.

This free course by Michael Pavlovitch is excellent to start with both apps.
https://gumroad.com/l/JsSEp

SnowInChina

musashidan said:

It's $90 for 3 months. It's dirt cheap. It's not as if you need it activated at all times.

Photogrammetry has its place, but it's not essential learning by any means. Top substance designer artists are using scan results(especially height/normal data) as reference, to great effect. Recreate the real surface and still have procedural power to create many, many variations.

really depends on what you are doing with it

if you make money using it, yes, thats pretty cheap

if you use it for personal stuff... costs add up

i have bought 3DF Zephyr

its not on par with reality capture, at least from the comparison i have seen, but it gets the job done. they also pump out updates frequently

if you need this for your business, i would propably go with reality capture, for everything else i would buy zephyr/photoscan or use the open source software renderhjs mentioned (or the free version of zephyr)

Benjammin

There is a free version of 3DF Zephyr. Its more or less fully functional, but you are limited to 50 images per model. You'll easily hit that limit with large/complex objects, but its enough to learn the process and do plenty of simpler things.

TTools

Agisoft Photoscan is only $179 for a standalone license, and if you happen to be a student, it is available in an education license to accredited educational institutions and it's students for only $59.  I don't think it's quite as good as reality capture in some aspects, but the price difference is exponential.

From their webpage:
Educational licenses (rehostable node-locked) and Educational Floating licenses are available exclusively to accredited educational institutions, their employees and students of such institutions. Only officially accredited educational institutions are eligible to obtain Agisoft PhotoScan educational licenses. This includes: universities, colleges, junior colleges, scientific and technical schools, vocational schools, and correspondence schools. Educational or Educational Floating licenses can also be purchased by students and officially employed teaching and research staff of officially accredited educational institutions. 
Any educational license prohibits all commercial uses of the software. Educational / Educational Floating and Stand-Alone / Floating licenses differ legally but are technically identical.

http://www.agisoft.com/buy/online-store/educational-license/

http://www.agisoft.com/buy/online-store/

radiancef0rge

no

tombig

Photogrammetry is quite simple in my opinion. There are many site where you can find info how is it done
http://svizze.com/articles/photogrammetry-quick-introduction/
I think you can learn basics in about one week.

haffy

I am about to try this photogrammetry thing out, I will set up a small studio for this and also capture stuff outdoors. I will mainly scan nature elements, like bark and rocks to use in zBrush. Will Photoscan work for me or should I absolutely go for RealityCapture?

And do anyone have tip on materialscanning process, for use in Substance Designer and maby a gumroadtutorial on this

oraeles77

haffy said:

I am about to try this photogrammetry thing out, I will set up a small studio for this and also capture stuff outdoors. I will mainly scan nature elements, like bark and rocks to use in zBrush. Will Photoscan work for me or should I absolutely go for RealityCapture?

And do anyone have tip on materialscanning process, for use in Substance Designer and maby a gumroadtutorial on this

my experience since I started this post is that reality scan is much much better than the others, and much quicker too.

sacboi

For what I do, possibly but on the other hand 'should' we?! well...in my honest opinion probably not.

Neox

you should rather focus on basics, composition, colortheory, game/leveldesign theory, design in general all that jazz that helps you build better environments for game. if you focus on a technicality like this you will be out of your job once photogrammetry is done automatically. and it will be done automatically, nobody likes cleaning scans, nobody loves doing retopo on mushy scans, all this shit that nobody enjoys and is super repetetive will be done automatically at one point in the future. it can't hurt to look into it, understanding it is certainly not harmful, but dont rely on it. if you can do great stuff without, you willmhave no problem doing greatvstuff with scan support.

Laughing_Bun

Reality capture is amazing. It is very cool and I'm a believer, but in a production environment it really does have limited application. You need the whole team to buy in on the process, most studios are run from creative directors and art directors so just making stuff from scratch is more appealing than going out and shooting real-world content then cleaning it up. For the right projects it's amazing, but I would never say it's manadtory learning or "the next big thing"

CreativeSheep

No.

musashidan

I would say it's more suitable for digital human/digital double creation.

The funny thing is, I've seen plenty of hand-built rocks, for instance, that actually look better than their real-world counterparts that have been built using photogrammetry. A case of art imitating life imitating art....

Another interesting occurrence that's come out of the scanned data craze is that Substance Designer artists are now using scanned data as reference to rebuild as a Substance material, rather than using the scan asset directly. This offers much more power and flexibility.

gnoop

It's not that much to learn actually.  It's rather a matter of having nice camera and certain shooting approach.

 Rebuilding scanned things  in Substance Designer looks a bit like a sport for a sake of sport  or to prove something.

   Anyway best looking  substances are usually a mix of scanned base  and procedural extra details and such approach is a huge time saver.

 

Besides there is Artomatix .   That's really next "big thing"   It could make countless variations from photoscanned sources in no time.   They  did a huge progress last year.  Their "content aware fill" node does wonders and their 3d preview is better and more relevant than SD one now.

Imo It would probably kill   Substance Designer in foreseeing future since after all those years   Allegorithmic  still couldn't make their soft truly convenient and easy to use

phaedarus

While photogrammetry will prove to be a valuable addition to your workflow, the main difficulty is selecting the correct hardware that will give you the most bang for your buck down the road.

Here are some of the technologies I’ve worked with that you’ll need to consider:

-          DSLR camera. I currently own an outdated crop sensor Nikon with 14 megapixel resolution. I use it primarily for texture sourcing. It’s bulky and awkward for photogrammetry but the results aren’t bad. I’ve thought about buying a used full frame camera like the D800 for higher resolution and the ability to shoot in situations where the light is not ideal but it’s still a pricey route to go in addition to being even bulkier and more awkward to use. In a professional studio environment however, they do give the best results; especially for turn-table captures.

-          Drone. I’ve seen some very impressive 3D models produced as a result of drone photography for architecture. Drones will no doubt be indispensable for large structures that would otherwise be difficult for ground photography to provide full coverage for. The problem with drones are the limited flight time, environmental hazards and the ever restrictive drone laws that are being passed which may soon make them impractical for use in urban environments. A good drone like DJI’s Phantom 4 Pro cost a tidy sum and if you lose it due to an accident or because someone thought it would be fun to shoot it out of the sky, you’re out a thousand+ dollars.

-          Laser scanners. I’ve kept track of both numerous kickstarter projects and conventional hardware - the most popular being Microsoft Kinect which scans depth on a surface rather than evaluate photographs. The major advantage of scanning an object is that it is mostly foolproof – little technical knowledge is required. If you can comb your hair, you can complete a full laser scan. This is possible due to the real time feedback of the scan on your device so you can preview for any missing details such as surface gaps that may require another pass with the scanning instrument. Unlike still photos, backgrounds in the scanning environment have no affect on the final result thus, saving a lot of tedium in the process. Laser scanners are more sensitive to lighting conditions than most other instruments when capturing surface texture and therefore are a poor choice for work outdoors. Consumer grade scanners are very affordable but suffer from low resolution and therefore not suitable for very fine detail capture. They also struggle mightily with wet surfaces. The cost of higher end scanners are in the stratosphere range and therefore, not really a consideration.

-          Smartphone/tablet. This is what I use for a majority of my work. The benefits of ultra-portability are fairly obvious. My own iPhone is several generations behind having just an 8 megapixel camera but it is adequate for obtaining reference information on colour, tone and shape when run through Agisoft Photoscan. The iPhone Xs has 12 megapixel camera which I expect would net you even better results. In addition to the device itself, there are third party hardware add-ons that aid in photogrammetry such as the Structure Sensor and EORA 3D. These add-ons are physically bracketed to the iPhone/iPad and through hardware in combination with specialized software, renders out point clouds in near real time right on the device’s screen. The scan resolutions are based on the device’s camera; the better the camera, the better the results. Currently, the resolutions are low due to compromises made for the real time element but as improvements are made to software optimization, it could well approach close to full frame DSLR quality in the near future.

 

As you can see, photogrammetry has branched out to different platforms; each with their own pros and cons. Experiment with what you already have and see what you can do. If you decide you want to upgrade to better hardware later, keep in mind that consumer photogrammetry is still maturing with no hardware standardization. That means you could spend thousands of dollars on DSLR camera gear only to discover a year later that a newer smartphone model or dedicated device will be able to output models of identical quality for a third of the price. It may actually pay to sit on the fence for a few years to see what the technology leads us before investing serious money into it.

gnoop

Many 15-20 mpix  DSLR cameras are not that much better for photogrammetry than iPhone dng saved shots done under perfect lighting.    But  since best photogrammetry is under cloudy lighting  it's always not enough light and too much noise usually for a smartphone.     I don't think it will ever change.

   In my own experience the resulting geometry is always marred with weird looking caverns and lots of geometry noise with both smartphones and consumer class drones like Mavic.

 I have $900 compact Sigma dp2 camera with Foveon matrix/Fixed lens.  It's  better in resulting geometry , cleaner and more detailed than  few Canon/Nikon DSLR  cameras around 20 mpix we tried.    I think visually the shots are closer to 40 mpix Bayer matrix camera

    It comes not without  disadvantages. It's slow as hell and you have to wait 2-3 min after each 6 dng shots, too narrow angled   so for a ground subject under your feet  you have to do a shot after just  20 cm shift  and works ok with no more than  100 iso,   struggles under dim lighting. 

   Still could be not so expensive choice

A good site to chose a camera

https://www.dpreview.com/reviews/image-comparison/fullscreen?attr18=daylight&attr13_0=apple_iphonex&attr13_1=canon_eos6dmkii&attr13_2=sony_a7rii&attr13_3=sigma_dp2q&attr15_0=jpeg&attr15_1=raw&attr15_2=raw&attr15_3=raw&attr16_0=32&attr16_1=100&attr16_2=200&attr16_3=100&attr126_2=1&attr171_2=1&normalization=full&widget=1&x=-0.5932333810195433&y=0.28601913939595586

AlecMoody

Photogrammmetry is a useful tool but it isn't a magic wand. Also, you will have way more success with the workflow if you are scanning things that are within your skill level to convincingly model and sculpt yourself. There will be cleaning, remodeling, etc, involved. I think I have done about as much scan based game modeling as anyone else and occasionally I do something with scans that is a massive time win. More often losing the ability to simplify the high res model silhouette ends up making low res work several times slower and there isn't much total time gain. This will depend heavily on what kind of content you are building.

If you want a good single camera scanning setup this is the formula:

*3-4 monolights with strip boxes. This is the right amount of diffusion without making the space un-navigable with too much softbox. You don't need powerful lights. I have alien bees b400's and I could go with less power without issue.

*Modern full frame sensor camera + quality lens. I'm on a D800E and a Sigma 50 1.4. Dynamic range is about as important as resolution. Many objects will have deep cavities and having tons of shadow detail without noise will let you get scan data out of them. For the modern sensors, I would prioritize dynamic range over resolution. All of the current gen full frame sensors are more than high enough resolution.

*Treat surfaces wherever possible with developer spray, baby/foot powder, etc. Shoot with a polarizer if needed (I never use a polarizer).

*Process in reality capture

*If you need a scale reference, a precise machinist ruler with very fine but clearly placed tick marks. The one I use has half mm steps and a matte aluminum finish. When I have tested my scale accuracy it's well within limitations of repeatable measurement from my cheap calipers

*I never use texture information from the scans. Surfaces are treated for maximum scan quality, materials are built very quickly in substance painter. This works best for my content but you might chose a different approach for other things.

Assuming your lights can handle rapid cycling without overheating and while staying consistent, you can bang out close to 1000 photos in ~30 minutes. Since you are using flash lighting you will be shooting at about f/11 and there won't be any camera shake. If you treated the surfaces ahead of time you will be rewarded perfect source photos. This approach is all about rapidly getting lots of top quality source images while being able to hand hold. Shooting with continuous lighting on a tripod is slow, prone to error (oops, you didn't wait long enough for the camera to settle before hitting the shutter release) and you are very limited in where you can get the camera (good luck getting above and below complex and large objects). Shooting handheld is faster, way more flexible, and with the right lighting you should have a 100% success rate on getting tack sharp images.

Here is a sample scan result. This was 297 photos and my time stamps say I spent 13 minutes shooting. I planned to remodel all of the plastic parts so I didn't focus any images on them, I only put effort into the cast aluminum.

Mark Dygert

is Photogrammetry the 'big thing' we should all be learning?

That's kind of like saying photos replaced textures when Max Payne came out in 2001. Or normal maps replaced hand painted details. It's all helpful sure, but it doesn't do everything that everyone needs. There are a large list of objects and styles that it doesn't apply to. Plus the costs and time involved, it can make it just as expensive and put enough friction in the pipeline to make it unusable.