Jason Devitt

823 posts

Jason Devitt

Jason Devitt

@devitt

Co-founder of Vindigo, Mr. Number, and Compound Eye.

San Francisco Katılım Temmuz 2007
490 Takip Edilen754 Takipçiler
Scott Ames-Messinger
Scott Ames-Messinger@scottmessinger·
@devitt This is totally random, but your CTO, Konstantin Azarov did a tech talk about LunaDb back in 2015 when he was at Asana. The video has vanished from the internet and I was trying to see if he had a recording or his notes. It was a really transformational tech talk for me
English
1
0
0
28
Jason Devitt
Jason Devitt@devitt·
@j___lambert @fdellaert Part of the genius of that paper is that it doesn’t require the NeRF geometry to be accurate to train stereo; it only requires accurate view synthesis.
English
0
0
1
72
Frank Dellaert
Frank Dellaert@fdellaert·
Question: is there any paper yet with a NeRF-based method that outperforms other methods *quantitatively* on multi-view depth estimation? Asking for a friend :-)
English
7
8
46
10.3K
Jason Devitt
Jason Devitt@devitt·
I'm at CES giving demos of @compoundeyeinc real-time 3D perception tech. We leverage automotive-grade cameras for on road and off road perception tasks. Find me at booth #61401 in Tech West at the Venetian Hotel Expo January 7th and January 8th. #CES2023 #AutonomousVehicles
Jason Devitt tweet media
English
0
2
7
1.3K
Jason Devitt
Jason Devitt@devitt·
@jon_barron @cut_pow Yes, but the diffusion model alone is used to inpaint the occlusion zones that are created when the frame is warped according to the depth map.
English
0
0
1
0
Jon Barron
Jon Barron@jon_barron·
@devitt @cut_pow It sounds like this approach is using a monocular depth estimation algorithm (MiDaS) that is unrelated to the diffusion model, am I understanding that correctly?
English
1
0
0
0
Jason Devitt
Jason Devitt@devitt·
This beautiful work by @cut_pow suggests diffusion models have at minimum learned to reproduce static monocular depth cues. Not clear if that's the same as learning 3D geometry. But if not, MiDaS has no 3D knowledge either; it uses the same cues. FYI @jon_barron
Dmitrii Tochilkin@cut_pow

1.2 It heavily relies on the quality of depth maps, and uses the assumption that SD has implicit knowledge of the scene geometry in an image. So therefore it can plausibly inpaint missing parts without explicitly knowing underneath 3D meshes of the scene

English
1
0
2
0
Jason Devitt
Jason Devitt@devitt·
@jon_barron @poolio @BenMildenhall @ajayj_ OK. I am really surprised that diffusion models usually produce images that follow rules of perspective. In some hand-wavey sense StableDiffusion has learned more about 3D geometry than it has about … hands.
English
0
0
0
0
Jon Barron
Jon Barron@jon_barron·
@devitt @poolio @BenMildenhall @ajayj_ I'd politely push back against both of those characterizations. We don't have any reason to believe that diffusion models have learned 3D representations, and we're getting consistent views here because we explicitly parameterize the scene *in 3D* with a NeRF.
English
1
0
2
0
Jason Devitt
Jason Devitt@devitt·
#stablediffusion is remarkable. The failure cases are as interesting as the successes. Prompt "photograph of a young woman looking through a telescope, professional" @karpathy
Jason Devitt tweet media
English
0
0
0
0
Kyle Vogt
Kyle Vogt@kvogt·
1) When I started @Cruise over 8 years ago, I was told self-driving cars would always be science projects. They were wrong. It was a grind, but we made the tech work. Now we’re building a business.
English
46
161
1.8K
0
V
V@M_Parabellum·
@devitt @TheAutonocast @TaraniDuncan @compoundeyeinc I wasn’t calling you guys out. It is a broader observation on this topic. Though you guys did hint at camera-only being a viable AV solution, but didn’t seem to address redundancy. I’m curious as to how your solution would solve depth perception in a snowy “whiteout” scenario?
English
1
0
0
0
V
V@M_Parabellum·
@TheAutonocast @devitt @TaraniDuncan @compoundeyeinc It’s funny how the lidar company CEOs never talk about the future being lidar-only or lidar being the end-all be-all… yet the camera vision companies place themselves at the center of the world at all times.
English
1
0
0
0
Michael DeKort
Michael DeKort@imispgh·
@devitt @TheAutonocast @TaraniDuncan @compoundeyeinc 2)For public domain cameras are not yet capable of being a main sensor unless you remove their weaknesses through micro, and usually useless, ODDs. Right now you need fusion of camera with IR, LiDAR with tracking & classification and imagining radar with classification.
English
1
0
0
0

Keşfet

@scottmessinger @fdellaert @compoundeyeinc @jon_barron @cut_pow @poolio @BenMildenhall @ajayj_