I am at a loss

[Rel]

I have been trying to scan an object for a few days now


but I can never get even a kind-of recognizable result. this would be my best:

this result was after manually masking each of the 222 photos myself. the "automagic" helper was almost useless, and the edge detection was frustrating.
but I'm not here to flame, or be unreasonable...

It seems like the more info I give the software the worse it gets. I tried giving it 1000 frames from a video, incoherent mess. and it cant identify the plate he is standing on as a flat surface, there are points going everywhere.



every time I would try and use more than like 400 photos on the normal setting for just sparce generation, my computer would freeze at one of the last steps after an hour of processing (I have 32 gigs of ram, 100 gigs of storage space, rtx3070) and all I could do was hard power off my pc to fix it.

everything I am describing I have tried multiple times with all the different settings and I cant come up with anything remotely passable.

the figure is a bit glossy, so that may be part of the issue. you can see it has way more points for the plywood than the model (not that that made the plywood very flat or defined in the final result.......)

I plan on trying again with some polka dot stickers but I have little hope at this point.

[cam3d]

Hey Rel -

Your subject is not 'photogrammetry friendly' - Smooth homogenous surfaces with few features to match between photo to photo need to be prepped with talcum powder or a similar fine dust. This dust adds lots of features which help with camera matching and surface extrapolation.

Your photos are all at the same elevation and with VERY high overlap. You would see much better results setting the camera at 4-6 different elevations with a bit less overlap. Everyone's hardware is different, so it will take a little while for you to find the sweet spot which works most of the time. As a general rule I shoot between 36-48 photos per elevation, and typically 6 different heights. YMMV.

The automagic helper needs particular conditions to be effective. I'd actually recommend instead using the mask by colour function and taking your photos with this in mind. https://www.3dflow.net/technology/do...for-turntable/

Video is weaker input data than photo. This is because of things like reduced resolution, interlacing, image stabilization and other post processing concerns. If you can take photos instead, take photos.

32GB is on the entry level end of the scale for photogrammetric processing, so if you want to use more images or higher settings, I'd seriously consider an upgrade.

100GB is not much storage space at all - I personally I like to have at least 250-500GB as a buffer, otherwise things can get a bit dicey.

Don't do the polka dots. There is no value in these for photogrammetry, they are for tracking handheld/robotized scanners.

[Rel]

Thanks so much for the information, I will make another attempt at some point.