I need your advices to build successful models with photogrammetry software. Tried Context Capture Master, Agisoft Photoscan, Zephyr and others

(Kemaleksen) #1

I’m a 3D animator and photographer. As photogrammetry covers them both I have been interested with photogrammetry software for a long time. I once had a gallery at hypr3d.com, used 123Dcatch, some software where you had to put your subject on a funny callibration paper, where you only had to shoot two photos and still have that Scann3D on my phone.

But the software that I tried in last years such as Zephyr, Autodesk Recap360, Remake and especially Agisoft Photoscan and Context Capture Master exceeded the quality of the previous software that I worked with. But still I don’t think my results are good enough to use for commercial purposes. I would be glad if you have a look at my sketchfab gallery: https://sketchfab.com/Kemaleksen/models

Photogrammetry software that I have seen have a simple workflow compared to my usual 3D modeling and animation software, but I wonder if I’m missing some little steps at the workflow or during preparation of photos. I usually use about 20-30 photos or 100-200 or sometimes several hundreds videoframes at photogrammetry software, don’t change focal length, but I use auto exposure, scale down images to about 1000 pixels at long side to avoid to wait forever.

Agisoft Photoscan doesn’t like my 17mm wide angle images and builds pyramids from photos of regular buildings that I shoot from below, so I have to crop my photos by 10% for it.

Models of Zephyr are usually not as good as of Photoscan (side by side tests at Sketchfab prooved that too) but at least Zephyr can handle wide angle photos.

Context Capture Master is slightly better than them, but sometimes worse, it’s a bit weird that you have to switch on that Context Capture Engine too before it could process images and you have to avoid using non-english letters for your files and directories, otherwise the model would be built anyway but you can’t open it. Sometimes it builds half of the model very tiny, only CCM does that, just scroll in my model to see it: https://sketchfab.com/models/7297e8344e9645b4a1f90012964798a1 .

As you might have seen at my gallery the models of the buildings (my favourite subject) that I capture look generally like damaged by an earthquake. I would be very pleased if I could get your advices to capture better 3D models. And maybe you could tell me where I shouldn’t expect more from current software technologies.

(Carowley) #2

Well… I am in the same boat as you. Autodesk Remake was a very good program before the ReCap changes. For close-range photogrammetry I was able to pull the best object models from Remake. Recap can still perform just as well but it is expensive in my opinion and it is cloud base. I recommend trying Recap but small objects take a while to process. You can clean and fill holes of the object within one piece of software simplifying the workflow. I believe you can do under 100 photos during the trial process.

I used a Nikon D5500 with the default lens 18-55mm lens. The mastodon and mammoth teeth were capture with this software and equipment. If you like the results that is the route I went. Hope this helps you I am moving toward aerial photogrammetry. ReCap 2018 can also do up to 1000 photos but it will be pricey to stitch without knowing the end result.

Hope I was helpful.

Chris

(Joel Whitton) #4

There are so many photogrammetry software at the market and you constantly trying to use new and new ones to find the most suitable program for youself… I still can not stop choosing, but if you need some advice from the person who tried different software PM me pls.

(Kemaleksen) #5

Yes, sure I would need your advice. I want to PM you but your Sketchfab account doesn’t have a Contact button.

(Joel Whitton) #6

i’m new on this forum, sorry, let me see why is not working. Sorry again

(Kemaleksen) #7

OK. If it’s not working would you please drop a message via my Contact button?

(3dsam79) #8

a few things that will help you achieve better results.

don’t resize your images, any downsampling you do will have negative effects on the quality of your mesh.
never crop your images. can’t recall why exactly, but i know it’s a big no-no for photogrammetry, possibly to do it modifying the exif data of the images, which the software needs for alignment.
take more pictures. for an average stand alone model you should be using around 100 images. the less you use the less accurate and more prone to errors and artifacts your mesh will be.
find a piece of software that works well for your pipeline. (personally, i use photoscan and realitycapture, depending on what i’m capturing and how much post-processing will be required.

most of the steps I’ve suggested will increase your processing time but are necessary if you want to step up your photogrammetry. however i will add that if you haven’t tried realitycapture already then definitely give the demo a try at least. its literally 10+ times faster than the other apps i’ve used. the only downer is the auto texture unwrapping function is a pain in the A to work, so whenever i use it i always unwrap the UVWs myself in 3dsMax

(Kemaleksen) #9

Thank you very much 3dsam79. I haven’t tried to shoot 100 photos for a model yet but I have tried to build models from several hundred video frames, mostly from interior walkthoroughs. Then for that purpose I probably need even more photos and without motion blur.

Resizing sure has a negative effect on mesh quality but don’t you think that 21 megapixels from my camera would be overkill and meaning infinite processing time?

I do cropping only when processing wide angle (17mm) photos at Photoscan. Somehow it doesn’t understand the wide angle distortions in a photo but thinks that the model is actually distorted. So it builds pyramid shaped models from ordinary buildings.

I read that Context Capture Master is actually the upgraded version of Reality Capture, and I didn’t have the opportunity to work with Realiy Capture anyway. Context capture has slightly better results than the other software I tried but not always. Sometimes it builds half of the model at a very small size, sometimes there is even a third part of it in micro size, only in Concept Capture Master.

Talking about interior photogrammetry, capturing blank walls, thin structures such as wrought iron furnitures and shiny parquet floors are always a problem for me. I probably need to wait for further progress of photogrammetry software.

(3dsam79) #10

sorry for the delay,

i would say no it’s definitely not overkill. I use 24mp images and for some projects i use several hundred images, never downsized(most studios use 50mp cameras, unless it’s a multicamera rig). the way i look at it, the more pixels you have and the more images, then the more accurate the results will be. even if i decimate the mesh i still use the high poly version to bake the normal maps later on. without a high poly mesh, the normals either have to be faked which isn’t ideal or u end up with a flat model.

as for processing time, it obviously takes longer to process more images but it depends what software you are using. If i were to process ~100 images in photoscan with settings on ‘highest’ it would take the best part of a day to process. the same project in RealityCapture would take about an hour.
I do however still use photoscan from time to time as i find it better for certain tasks.

a 17mm lens is very extreme for photogrammetry, the images probably need ‘correcting’. you can get some software that removes the fishbowl effect from wide angle lenses. not sure what it is exaclty but i know it’s recommended/required if you’re using a gopro for example (which i think has something like a 18mm lens too). The ideal focal length is 50mm.

shiny surface don’t capture well, neither do blank featureless expanses, like plain interior walls.
you can sometimes use dulling spray on shiny object to kill the reflections but a floor is just too large imo. for the walls you could try adding some unique identifying markers. some people use post-it-notes or stickers, adding them to the walls to help with the capture. they will of course need removing/fixing from the model afterwards.