One of the good things about working at a uni is educational pricing on software. Playing with Agisoft Photoscan at the moment. Better models than 123D Catch :D My 3D sculpting skills are still rudimentary, but I managed to smooth this model enough to look reasonable up close.
That looks really good. How many photo's did you have to take?
This one was about 70, but it was an old image set and I only had jpegs. The sculpture also has a few areas with very little detail which caused a few problems and had to be smoothed out. Still learning Photoscan but it's a great program for the price. Results are much better than 123D Catch and for some objects it's better/more efficient than our NextEngine laser scanner. the texturing is pretty good too. Working my way up to a building, or at least enough of one to include in a scene... or a detailed tree trunk for a foreground object.
I looked at the images before reading the text.
So I thought you created the 3d object file and then you made a real 3D render of it :P
Then i read the text ::)
But anyway good use of 123D catch
David
How is the quality of the UV mapping and the textures compared
to 123D catch?
Please elaborate, BB. You know this topic is too interesting for just a blurb. :)
Quote from: j meyer on May 31, 2014, 11:02:00 AM
How is the quality of the UV mapping and the textures compared
to 123D catch?
http://bigben.id.au/demo/mini-mali2.tif (http://bigben.id.au/demo/mini-mali2.tif)
1 of 4 (I added alpha channel for TG)
Pretty good. I had to clean up the model and import it back in so there are still a few odd bits from my average cleaning job. (Being able to export the mesh for cleanup and then import it back in for texturing is really useful)
I find this program much better than Catch as you have total control over texture size (size of image and number of images created) and can mask out sky/people/traffic. The only thing you can't do in the standard version is define matching points between images and set the scale of the model.
I tried to do this model in Recap360 (the updated version of Catch) but half the uploads timed out and a full res file requires a subscription. I gave up on the uploads after I had a dense point cloud in Photoscan.
http://www.youtube.com/watch?v=qNLzot1o_1c (http://www.youtube.com/watch?v=qNLzot1o_1c)
https://vimeo.com/40602544 (https://vimeo.com/40602544)
Quote from: TheBadger on May 31, 2014, 07:44:50 PM
Please elaborate, BB. You know this topic is too interesting for just a blurb. :)
Moving beyond 123D Catch http://www.planetside.co.uk/forums/index.php/topic,14945.0.html (http://www.planetside.co.uk/forums/index.php/topic,14945.0.html). I didn't render that model too close because there were spikes I hadn't fixed. Reprocesing other photos from around the same time.
[attach=1]
Full price of the standard version is just a bit higher than I'd like to pay without testing saving/exporting, but the education price was irresistable. The workflow is relatively simple although you need to learn a little bit about the process to get the most out of it... and of course a computer with grunt helps, especially the creation of meshes which is RAM sensitive. That said, I've only got 8Gb at home on a 4 year old computer.
I'm also testing it out at work and it's giving our NextEngine laser scanner a run for its money in terms of resolution and processing time for a range of subjects. Texturing on the NextEngine sucks big time, but I can now create a low res model in Photoscan, align the laser scan to that and import it into Photoscan for texturing.
There's a functional demo version.
http://www.agisoft.ru/products/photoscan/standard/ (http://www.agisoft.ru/products/photoscan/standard/)
and I use Meshmixer for mesh cleanup (free)
http://meshmixer.com/ (http://meshmixer.com/)
Thank you once again to introduce your experiment !
Thanks for this tutorial
A concept test - high res tree trunk for a foreground object. This one was created using a GoPro for the photos. NB. not really a suitable camera for this (FOV is too wide) but it worked OK. Also should have shot it on a cloudy day and cleaned up the mesh a bit, but you get the idea.
Both models are 500,000 faces with 4, 4096x4096 textures.
That's really promising and good info.
Another software that can be of help cleaning up meshes
is MeshLab,also free,in case you don't know already.
Hi
another workflow that I have seen that uses this thinking and methods also uses retopo on the scan. So one example I saw was the model gets taken into mudbox (after its all fixed up (mesh lab and such)), and then the scan model is sculpted on and repainted. And finally they did retopo and exported various maps.
I have still never did a retopo on anything though. So no real idea how much time that can add to a project.
Michael - Don't know about mudbox,but in ZBrush this can be done within
a few minutes( the retopo+mapping+and texture projection).
The Ten24 guys do their stuff like that.
Even the clean up can be done in ZB btw.
^^
Ahh, well, that is new info for me.
I just never read a description of that part, or saw a real time of someone doing it. I thought like everything else it must take ages ;D. Well good! Then this finally sounds like a practical and complete workflow to me.
I toyed with 3d catch on my phone. But Using a Iphone to take serious photos feels unnatural to me. And I did not like the 3Dcatch web browser plugin very much so never broke out my real camera for this.
I will take a look at the new service/soft that Ben linked.
Mud has retopo now, and there are also plugins. So , Probably should just start with a cola can or something. I think I have had my fill of bitting of more than I can chew... Finally. ;)
Meshlab is indeed useful for a lot of things, particularly stitching multiple point clouds together and surface reconstruction of point clouds. Meshmixer isa bit more like ZBrush and Mudbox with its sculpting/smoothing tools (but it's free) and on the model repair side it's useful to get visual indications of where the mesh errors are. ZBrush still does my head in a bit, but in may have to keep learning that, if only for tidying up the texture maps at the end, but for TG this is not a critical step.
You can also export the cameras from Photoscan. Not sure if ZBrush can use these but I think Mudbox can.
The elephant was technically a difficult subject because it's very smooth with large areas of even colour and I had to do a fair bit of editing of the mesh. The tree on the other hand was a lot easier, and has Had no editing of the mesh. erros introduced by the use of an extremely wide angle lens were removed from the point clouds before generating the final mesh.
I'm finding this is meeting a number of people's needs at the uni, and one of the things everyone mentions is"fancy visualisations" of the objects/scenes they want to capture for which TG is pretty good at. Yes, it looks like I can finally get paid to do this stuff at "work" ;D
If you're already using 123D Catch to create objects I'd say that the standard version of Photoscan is a worthwhile investment. We're using Canon 5Ds but that's because we have them. Any camera with a good lens that can save RAW images at 12-18MP will do. The canon S110 for example is used in drones to do stuff like this: http://m.youtube.com/watch?v=NuZUSe87miY (http://m.youtube.com/watch?v=NuZUSe87miY) (Software for that was Pix4D, but it's the same principle)
One last sample. Reprocessed the images as seen in this cheesy video https://vimeo.com/66036184 (https://vimeo.com/66036184)
This was originally done before we had any mesh editing tools so to get a good model we had to keep adding control points until it worked. The model for this render was processed entirely in Photoscan. Textured objects like this work really well with photogrammetry.
Yup,Meshmixer is a nice free app.I remember that many people were quite
exited about the mesh stitching back then.
ZB does your head in? That is something I really don't understand.I got my
copy right after I tried to learn blender for several months,which did terrible
things to my mind.Then I got ZB and learning that was like a therapy and
my mental health was restored almost immidiately. ;)
No,ZB can't use exported cameras.
I can see this technique being of use to many scientists.
Good luck for getting paid for this.
Why do you need RAW images?
Quote from: j meyer on June 02, 2014, 11:29:24 AM
Why do you need RAW images?
JPEG compression artefacts and ISO noise are falsely interpreted as image detail so to get the best quality mesh you need the cleanest possible images. As low an ISO as practical and TIFFs from RAW images, but you can still get pretty good results with JPEGs
300k model of the first elephant: http://studio.verold.com/projects/538c4d646d602c02000003ed (http://studio.verold.com/projects/538c4d646d602c02000003ed)
Only got 2 of the set: http://www.zoo.org.au/150/mali-sculptures (http://www.zoo.org.au/150/mali-sculptures)
Looks good Ben.
Hey ben,
What kind of art/history/antiquities access do you have at your school? ;)
A lot more than I know about. I'm at Melbourne University, working here: http://digitisation.unimelb.edu.au (http://digitisation.unimelb.edu.au). There are a number of cultural collections within the university and then there are departments like Archaeology that send researchers all over the world. I've just done a couple of samples for some guys off to Turkey this week and they're keen to make models of things they can't bring back.
Here's a brochure for our collections
http://www.unimelb.edu.au/culturalcollections/collections/melbunicollections.pdf (http://www.unimelb.edu.au/culturalcollections/collections/melbunicollections.pdf)
Gotta get in there!
Ben i hope you don't mind it is something related that i thought could fit to your thread.
http://www.theastronauts.com/2014/03/visual-revolution-vanishing-ethan-carter/
Don't mind at all. It points out the inefficient textures that Photoscan creates. I'm still learning the workflow options, but as I don't have a lot of other 3D modelling software (or experience with them) I'm looking at simpler options. One nice thing with Photoscan is that you can export point clouds and models at various stages of the workflow, edit them and reimport them to continue the workflow. I think you can update the UV mapping in another application and then import it back in for the texture creation step.
"export point clouds" Have you tried it and if yes with which software Ben?
For point clouds, I'm actually finding Photoscan to be pretty useful for selecting points easily. The selection tools in many other apps freeze up once the point cloud gets really big. For objects, Meshlab can be useful for creating a mesh from an exported point cloud. Poisson surface reconstruction with parameters 11,9,X,1 where X (samples per node) is based on the desired face count for the mesh and the number of points in the point cloud. X = 2 x number of points / number of faces. I follow this up with a quadratic edge collapse decimation to reduce the face count, incorporating planar simplification. Then import that mesh back into Photosan and do the texturing.
The other thing Meshlab is good for is editing meshes created with Photoscan. Arbitrary meshes have larger faces around some edges which often pick up texture from the distant background (e.g. sky). These can be selected in Meshalb by selecting faces by edge length (they get larger as they get closer to the edge). Alternatively, Meshmixer is handy in this instance in that you can easily select an edge and then grow the selection away from the edge.
Thanks for the detailed answer Ben :)
Quote01
CLASSICS AND ARCHAEOLOGY GALLERY IAN POTTER MUSEUM OF ART
Pre-classical to classical ceramics,
coins and manuscripts
Location Swanston Street, Parkville Campus Access Tuesday to Friday 10am to 5pm; Saturday and Sunday 12noon to 5pm
Contact email potter.info@unimelb.edu.au; telephone (03) 8344 5148; fax (03) 8349 3518 www.art-museum.unimelb.edu.au
The geographic focus of the Classics and Archaeology Collection is the ancient Mediterranean world and the adjoining regions of the Near East. Chronologically, the antiquities span several millennia and are testimony to many significant human achievements. Among these are representations of early writing systems, items of ancient maritime trade, artworks of the Near East and classical worlds, coins and illuminated manuscripts.
Sounds like they would have a statue or two of some gods? Athena and neptune? If you get into it, see if they have anything! If you need to keep working on this and you need subjects ;)
Don't even have to go inside ;)
http://sophieinmelbourne.files.wordpress.com/2013/04/ian-potter-museum-of-art.jpg (http://sophieinmelbourne.files.wordpress.com/2013/04/ian-potter-museum-of-art.jpg)
Oh man! Go inside go inside!
This is great... Must play with 123dcatch a bit more now!
:)