Tweet
I am a VFX director working on a project. I shot a bunch of scans on Sunday and have spent the last 5 days trying to get working models out of it, but have had no luck. I basically just followed the tutorials on 3DFLOW. I input the photos, mask the subject, after it builds a sparse cloud, I tell it to build dense cloud, and then tell it to create mesh. For my test shoot I shot several poses and several outfits with 2 different people using a RED camera shooting 6k 24fps. I then time stretched the footage 400% to reduce it to 6 images per second to get a nice supply of photos and angles. The sequences are roughly 500 frames long. I lit them with quasar tubes, daylight LED's, and raw daylight coming through a big window to get the most light. There still wasn't enough light, so the shutter speed couldn't be that high and the lens couldn't be f22. I realize that's part of the problem right there but I didn't have more light at the time. Also the lens was a 24mm, so I couldn't fit the subject in the whole frame. I spend several days converting the footage and running tests but it didn't work. So I shot another test with my iPhone and made sure the background had a lot of features the camera could see and did it in daylight on an overcast day. I still can't get any usable geometry. I'm hoping that the RED footage I shot was not wasted because I don't want to have to get the talent and rent lights again. What can I do to save my footage and get some working models for my project. It took me roughly 8hrs to mask 500 photos and I don't think I can waste more time running tests. So I'm wondering what I'm doing wrong and what I can do with the footage I have.
-
-
Hi there! - I know you've had feedback through Discord and our Help Ticket service, but I figured for anyone else reading this there are some points to be aware of:
1. Humans are always moving, even just a little bit of movement will cause problems because they will not be the most stable thing in the scene (normally floor and walls), and thus suffer from misalignment.
2. The clothing the humans are wearing is relatively featureless - This makes makes tracking extra difficult.
3. 24FPS for video capture is very slow, considering how fast and close you're moving in relation to the subject - This will invite motion blur into the mix (no fun).
4. Both shoots are done indoors, which means you have less than ideal lighting conditions, and subsequently your camera settings won't be optimal (f/stop too low, ISO too high etc...)
It's not impossible to achieve a body scan with this kind of setup, it's just very very difficult to get good results, especially because your talent is the one variable you can't fully control (they ALWAYS move - I've had these headaches myself too many times). I can't stress enough: If the human moves during capture (and they WILL, even just breathing or blinking) there's not a lot you can do, other than manually fixing up models in sculpting software.
Bonus points:
1. 24mm is a great focal length for all sorts of scans - I generally wouldn't go any lower because I like to avoid distortion in my lenses.
2. Camera rarely needs to be F22 (unless you're shooting in a controlled studio environment with really powerful lights) - f11-f16 is the 'standard' sweet spot. Camera DOF calculators are super handy for figuring out what you really need.
- Disclaimer - I'm just repeating myself for the benefit of those coming to the forum
-
Thanks cam3d, I am always up for more advice and tips
Comment
-
Here's a setup you would need for scanning living subject matter, it snaps every angle at the same time.
Pricey to own or find a place that provides the service
Comment
-
Yup, not the cheapest thing in the world. I can personally recommend ten24 in the UK and Avatta in Tokyo, Japan. Setting up those sorts of systems is a real art!Comment
Comment