Mika SaastamoinenIn the forest, a boulder has an important message written on it... 😉 #NeRF #render made with #Nerfstudio
Paul Melis@miksi What camera did you use? So far I've not been able to get them as "splotch-free" as yours...
@paulmelis I just use my regular Android phone, a OnePlus 8T, taking video at also very regular Full-HD (1080) resolution. I think the blotchy "clouds" and "fuzzy areas" are an artifact of not having enough image/video coverage of those areas. Not enough image data for the NeRF algorithms to work on. In this case I also had "cloudy" and fuzzy areas but kept them mostly out of view by adjusting the render camera path and camera angles just so... 😉 Also I think taking video of the subject is much easier snd quicker to capture, than taking individual photos of it. You'd need to take hundreds of photos ideally, very time consuming captured individually, but a couple of minutes of video gives you those hundreds of image frames for a lot less effort. Video also provides "good data" i.e. coverage of the background scene while you're walking around the subject, reducing those fuzzy/blotchy areas.
@miksi Thanks, my latest tests where indeed using video as input, from an iPhone 13, for the reasons you mention. How many frames did you use as input ("hundreds")? And this is with the default Nerfacto model?
@paulmelis With my source videos, typically 2-3 minutes long, the ns-process tool usually extracts about 300 images from the video. On occasion I've tried upping the frame count to 500-600 but haven't seen much improvement, and often ended up running out of memory. :) About 300 frames usually works fine for me.
@paulmelis ,,, and yes, default nerfacto model ... 😁