Any thoughts on this? So far all the open-source photogrammetry software has been clunky at best, but this looks pretty legit and comparable to programs like Agisoft's Photoscan. https://alicevision.github.io/#meshroomhttps://www.youtube.com/watch?v=v_O6tYKQEBA
>>645750>not just using VisualSFM, constructing from point cloud in MeshLab and baking in BlenderThat stuff somehow manages to look more clunky
>>645750I've been using that almost a month, it is easy af and throws great results but it is slow and it has no documentation, trial and error is a pain in the ass with that
>>645759>a frankensteinian puzzle of 3 different applications instead of a single streamlined package where each of those operations are done as nodes>>645772It does look slow in the video, though from what i've seen photogrammetry is generally a slow process. And yeah some more documentation would be nice. Hopefully more people will be making tutorial videos like this in the future - https://www.youtube.com/watch?v=k4NTf0hMjtY
>>645796>nodesI personally hate nodes, but if you love nodes, Blender lets you use them for materials, animation and compositing.Good luck actually modelling anything with nodes though. That's what Blender is absolutely amazing at. The sculpting is also becoming pretty nice.
>>645796I've been using arround 50 pictures 4k each and take arround 90 mins, some results
>>645800I've been using blender for nearly a decade now, there's no need to shill it :-)>>645801Not bad, it's better than what i got out of david-laserscanner and a line laser many years ago. Do you think the camera resolution matters much? Like i'm thinking that if you include enough sharp and zoomed-in (macro) pictures you can get high detail with any modern >5Mpx camera
>>645808All these are from upscaled videos from youtube, without that the output lose too much detail, some objects need more pictures if it has multiple objects, arround 100, deepmap x8 gives better results like that Gollum or the Star Wars soldier, I change also featureextraction if the program fucked the cameras positions, but sometimes is immposible
>>645815 ContAnyhow the results are just references for retopology, the ouput mesh is shit always
>>645808And too much pictures sometimes fuck the cameras
>>645862Not always work that well, the program was unable to get the fokker from this vídeohttps://m.youtube.com/watch?v=nXG36W5E5tk This is one of the ships that used Columbus, the output isnt that good neither.And I was unable to get the cars from the vídeos of this guy neither, but it was one of my first trieshttps://m.youtube.com/user/TheSmokingTire And I've deleted lots of fails
>>645874Maybe the results are caused by the lo resolution of the pictures taken from the video. Meshroom downscales pictures using a factor. I dont remember what node had the parameter, but you could try decreasing it.
>>645881Many things can affect the program, sometimes the lighting makes it blend with the background, you can miss small details that are core for the model, sometimes it takes too much information where you do not need it, extending the resolution works only sometimes, the main problem, if you manage to align the cameras, is that you cann't trim the point cloud before the depth map and reconstruction
>>645883Yeah, i'm guessing that a turntable will get you the overall shape, but not the level of detail you need without using some really crazy resolutions and computing power since the algorithm treats the background and object with equal interest. Close-ups are the thing that's going to yield the best detail.
>>645910Nope, I tried 4 times, 3 of them trimming the video and I only got the pedestal sadly, I saw the shape of the plane only in the basic pointcloud just like in the turnarround of the cars, that's why I think that should have a mask option or a way to trim the pointcloud, 20/30% of the time fail or throw shit, I don't know if could work if you modify that externally with CloudCompare ....
If a model is all in quads/tris but the shape of those quads are F U C K E D, will the final result be okay for a games engine?
>>645925wrong thread fuck
>>645801I don't think its the quality of the individual image (4k) as much as it is the number of different pictures of the object
>>645933I haven't been able to put more than 120 pictures and get good results, the program start fucking the positions of the cameras, maybe is my computer idk, and the times get much worse ...
>>645800>good luck actually modelling anything with nodesHoudini would like to have a word with you.
>>645800>I personally hate nodesNGMI.
>>646154Blender was just like that 3 years ago, Gimp still is ...
>>646160Nvidia are such scumbags, i thought we were finally getting away from this shit when OpenCL started to work in Blender.What hurts the most is how they state they are not only unable, but unwilling to change this... Hopefully someone comes along and adds at least CPU support, however slow it may be.
>>646184>Blender devs implement support for CUDA>Blender devs ignore FBXHm...
>>646198They don't use fbx, the developement was paid externally, anyhow they still support that and there's another exporter made by the comunity, they only develop shit for their open movies, that's how blender works
>>646198They don't ignore it exactly, Autodesk changes the FBX spec every year just to fuck up any progress made by other apps to support it.https://youtu.be/qJEWOTZnFeg?t=2813
>>646270>Autodesk changes the FBX spec every year just to fuck up any progress made by other apps to support it.lolwut? All of my programs work fine with FBX year after year.
>>646270All your autodesk programs?
>>646341Houdini, ZBrush, Substance Painter, and Unity. I try to stay clear of Autodesk's greedy business model.
>>645750There's been a lot of cool free stuff coming out lately. I would love to try this out in conjunction with materialize. Maybe that will be my project for the day.>>645801Pretty cool stuff, anon, although that F-4 looks like its seem better days, lol.
>>645750Is there a good procedure on terrain 3d scanning? I want to scan my house + yard. If I remember it correctly you needed to use some kind of geodetic instruments or something like that