Recent advances in 3D content understanding
ai.facebook.com
Recent advances in 3D content understanding
1–10 of 17 posts
Re: Recent advances in 3D content understanding
#2Kinda boggles my mind that there hasn't been a stronger push for hardware support of 3D imaging in phones. It clearly provides more useful information for analysis. Even when we look at an image in a photo we map it to a 3D projection in our mind.
Re: Recent advances in 3D content understanding
#3Kinda boggles my mind that there hasn't been a stronger push for hardware support of 3D imaging in phones. It clearly provides more useful information for analysis. Even when we look at an image in a photo we map it to a 3D projection in our mind.
Phones with multiple camera are starting to become common. Time of flight sensors that use projected infrared patterns aren't trivial to out in a phone so the demand needs to be there. Still, there have been tablets that have integrated Intel's depth cameras.
Re: Recent advances in 3D content understanding
#4Kinda boggles my mind that there hasn't been a stronger push for hardware support of 3D imaging in phones. It clearly provides more useful information for analysis. Even when we look at an image in a photo we map it to a 3D projection in our mind.
There have been a few phones which provided that, notably the HTC Evo 3D. It had dual cameras and an autostereoscopic (glasses-free 3D) display.
The value provided was, IMO, quite minimal. It was not easier to use or better in any user-facing discernible way, the 3D stuff just felt like a gimmick. 3D photos taken by the phone and viewed on its screen did not feel more lifelike. The color depth and image quality was poor even by the standards of other phones of its era.
While it was very cool and felt very futuristic, it did not feel worth the cost.
Re: Recent advances in 3D content understanding
#5Kinda boggles my mind that there hasn't been a stronger push for hardware support of 3D imaging in phones. It clearly provides more useful information for analysis. Even when we look at an image in a photo we map it to a 3D projection in our mind.
Think about the data explosion when all our photos and video store the whole light field.
A 2.5D model is probably less data?
Re: Recent advances in 3D content understanding
#6I do photogrammetry as a hobby, and would love to see more RGBD cameras on the market. I've even considered hacking together my own... anyone have pointers for cost-conscious options?
A few of the models I've shrunk down and posted: https://sketchfab.com/darkphibre
Re: Recent advances in 3D content understanding
#7Kinda boggles my mind that there hasn't been a stronger push for hardware support of 3D imaging in phones. It clearly provides more useful information for analysis. Even when we look at an image in a photo we map it to a 3D projection in our mind.
Phones with multiple camera are starting to become common. Time of flight sensors that use projected infrared patterns aren't trivial to out in a phone so the demand needs to be there. Still, there have been tablets that have integrated Intel's depth cameras.
Nit: Projected patterns are structured light, not time-of-flight. As far as I’m aware (would love to be wrong!) you can’t do ToF with a traditional CCD or CMOS sensor and resolution is invariably woeful.
Re: Recent advances in 3D content understanding
#8The videos showing the algorithm in practice are really nice demos.
I’m curious how big of a step forward this is from the previous state of the art, and at what computational cost.
Also curious if the technique scales well with multiple cameras with overlapping fields of view. That is to say, I assume accuracy can be increase through sensor fusion in the basic sense of averaging errors, but actually molding a cohesive 3D view of a 360° environment and understanding that an object at the end of one frame is the same object from a different perspective at the end of another camera frame.
Obviously this seems like it should be extremely useful for AutoPilot. Compared to the relative inaccuracy of the positional information of adjacent cars on the AutoPilot guidance display that we have today this seems like a big step forward.
I think it’s interesting how the RNN is identifying specific types of objects and then depth mapping them. I assume it can’t just depth map the whole image without that first classification step? I’m thinking like for the Smart Summon application where depth mapping everything around you is pretty crucial and obviously not entirely working at this point.
Re: Recent advances in 3D content understanding
#9I do photogrammetry as a hobby, and would love to see more RGBD cameras on the market. I've even considered hacking together my own... anyone have pointers for cost-conscious options? A few of the models I've shrunk down and posted: https://sketchfab.com/darkphibre
The new Azure Kinect is really impressive.
I've used the og Kinect, Kinect v2, and intel realsense d435 and it was much more accurate then all of those.
Re: Recent advances in 3D content understanding
#10It really seems like we should be close to having robots that can navigate a room.