Estimating Camera Motion from a Single Motion-Blurred Image

71 smusamashah 20 3/28/2025, 6:07:39 AM jerredchen.github.io ↗

Comments (20)

bsenftner · 33d ago
Visual effects, VFX, has been doing this type of work for decades. The guy that I know doing this the best, Eugene Vendrovsky ended up at Nvidia as a robotics computer vision division director. Back in the early 2000's he had all kinds of camera motion recovery algorithms we used in the Tracking department for use on feature films like the live action Scooby Doo, Riddick and the Narnia films. Eugene retired recently, after a very productive career.
anovikov · 33d ago
Can't upvote this enough. Finally a piece that's not about LLMs or how are they going to ruin the world.
fortran77 · 33d ago
And not about Rust, either!
damnitbuilds · 33d ago
And no cod psychology about how to live a better life, either!
dylan604 · 33d ago
or worse, make the world a better place
porker · 33d ago
Going to read this in depth. I have been on and off trying to find a way to distinguish between in focus, motion blur (camera shake) and out of focus. Motion blur is surprisingly tricky to distinguish using a computer from the others. I can glance at a 2D Fft plot and tell, but the computer - not yet.
fp64 · 32d ago
Blind Deconvolution is around for a while and from the estimated PSF you can gather the shake pattern (or tell whether it’s just overall out of focus). It just never worked good enough in practice but I remember some impressive results already a decade ago
porker · 32d ago
Thank you, blind deconvolution and PSF sound like what I have been scratching around the edges of in my experiments, without knowing the right terms to search for to discover prior work. I shall dig into the literature!
InDubioProRubio · 33d ago
Three days after posting- gets inverted and reused as a directional camera motion blur shader ever after
DrNosferatu · 32d ago
Doesn’t the Point Spread Function already achieve this?

https://en.m.wikipedia.org/wiki/Point_spread_function

drsopp · 33d ago
Isn't a logical next step to extract the depth field? Possible?
tetris11 · 33d ago
Isn't the depth decoder part of the processing already?
drsopp · 33d ago
I should have read the abstract.
skywal_l · 33d ago
Right there in the abstract:

"Our approach works by predicting a dense motion flow field and a monocular depth map directly from a single motion-blurred image"

wiz21c · 33d ago
I'm not familiar with all of this but is there then a tool to remove the blur ?
atoav · 33d ago
Deconvolution can be used for this, also see this currated list of resources: https://github.com/CVHW/Deblurring
wiz21c · 27d ago
This list is so big and goes so much in the past...
xattt · 33d ago
That’s for the next grad student to solve.
esafak · 32d ago
Yes, those have existed for decades, with various degrees of success.
Imustaskforhelp · 33d ago
Can this be used in LLM's like gemini which don't really have a motion idea but rather they just take things frame by frame and so can't really understand motion b/w them