Skip to main content
Conference on Computer Vision and Pattern Recognition

Conference on Computer Vision and Pattern Recognition

ConferenceWikipedia

Search complete. 12 mentions across 8 episodes found for "Conference on Computer Vision and Pattern Recognition".

Nov 20, 2026

George WolbergGUEST
21:04
And then furthermore, we can take these 3D edges and these 2D images and have them go through a 2D image to 3D model registration.
George WolbergGUEST
21:14
So some of that work is best described in a 2005 CVPR paper by Professor Stamos and Lingyun Liu.
George WolbergGUEST
21:22
And Lingyun is now here working at Google.
George WolbergGUEST
21:26
And basically, what that does is says that from all of these images, there are a subset of images from which you can extract 2D lines and match them to the 3D lines here so that you can recover the camera positions directly from that.

33 MINS LATER

George WolbergGUEST
54:42
No, no, we pick six here and we try to find six here and we keep, it's kind of like a brute force, it's iterative, finding six and six.
George WolbergGUEST
54:49
And it's actually the 25th anniversary of the introduction of RANSAC, which I think was 1982 or maybe 1981.
George WolbergGUEST
54:59
And there was a venue at CVPR last year which talked about the advances of RANSAC.
George WolbergGUEST
55:04
But I would direct you to how RANSAC works, those papers on how it works.
Yasutaka FurukawaGUEST
31:30
So, I've shown you, like, pictures of movies.
Yasutaka FurukawaGUEST
31:33
And of course, people are interested in asking, so how are they accurate? And our, so fortunately, there are some quantitative evaluation of multivisory algorithms, which was our published last year, CVPR.
Yasutaka FurukawaGUEST
31:44
So, basically, provide some data sets, then people run their own algorithm, then submit results.
Yasutaka FurukawaGUEST
31:51
They're going to evaluate your accuracy of the results, compare with the ground truth, and show you the how much accurate your results are.
Gowthami SomepalliGUEST
22:40
Okay.
Gowthami SomepalliGUEST
22:42
So that concluded the first paper, which we presented at CVPR.
Gowthami SomepalliGUEST
22:50
And so basically, we had this mainly from our analysis.
Gowthami SomepalliGUEST
22:54
What we found is there seems to be two factors which are strongly responsible for memorization.
Harry ShumGUEST
57:40
I don't know what is the latest conference coming up.
Harry ShumGUEST
57:42
When is the CVPR deadline? It just passed.
Harry ShumGUEST
57:45
The CVPR, if you type CVPR now versus CVPR later, and which CVPR you're looking for actually can be very different.
Harry ShumGUEST
57:53
So this is time dimension here.
Harry ShumGUEST
57:56
Certainly, this is the location in the spatial dimension here.
Damian BorthGUEST
9:00
So there are two different things how you could think about that.
Damian BorthGUEST
9:02
We had one paper that was published at Computer Vision and Pattern Recognition Conference, CVPR, where we had a very particular use case for remote sensing data.
Damian BorthGUEST
9:12
So satellite data and so-called foundational models that are able to analyze satellite data.
Damian BorthGUEST
9:19
And we now took our approach, trained it on the entire remote sensing neural networks that we could find on platforms like Hugging Face and, you know, had the pre-training on computer vision models and took this to generate foundation models in remote sensing.
Damian BorthGUEST
10:50
And now please predict checkpoint number 10.
Damian BorthGUEST
10:53
So you kind of jump ahead in transfer learning or fine tuning.
Damian BorthGUEST
10:58
For the numbers, the CVPR paper showed that we can be we can train comparable remote sensing models with 20 times less compute needed as compared to state of the art models trained from scratch.
Vit LyoshinHOST
11:10
Wow, that's amazing.
Kim IsenbergHOST
1:14
Jiangming holds a PhD in electrical and commuter engineering from the University of Illinois, and before Xpeng, he worked at Matter at Cruise.
Kim IsenbergHOST
1:25
He reports directly to XPeng's CEO and he is just back from CDPR, the world's top computer vision conference, where he laid out XPeng's roadmap for what the computer company calls physical AI.
Kim IsenbergHOST
1:38
Where he laid out XPeng's roadmap for what the company calls physical AI, XPeng describes itself as an AI company that happens to build cars and that's exactly what I want to get into today.
Kim IsenbergHOST
1:51
Jeremy, welcome and thank you for making the time.
Xianming LiuGUEST
1:57
I mean, the warm introduction makes me feel it.
Kim IsenbergHOST
2:01
I don't often have the pleasure to meet such great researchers, so it's a pleasure for me to meet you finally in person.
Kim IsenbergHOST
2:07
So as I was saying, you just came back from CVPR, the world's top computer vision research conference, right? Basically, the frontline of teaching machines to see what's the one idea from there that will actually change what Xpeng ships in the next 12 months.
Xianming LiuGUEST
2:24
Yeah, so we started working on something called VLA 2.0 from last year.
Adrien GaidonGUEST
40:44
Again, this is our network, better than the best self-supervised, better than the best semi-supervised or fully supervised, right? LiDAR, you don't have LiDAR points in the sky, so you have some issues with that, right? So that's also one reason.
Adrien GaidonGUEST
40:59
This is like last year's CVPR, no, two years CVPR.
Adrien GaidonGUEST
41:05
So just for your information like that, That versus that.
Adrien GaidonGUEST
41:10
The pace of the field is crazy.
Anastasis GermanidisGUEST
57:42
Like, are we actually going to be able to, you know, do we have the resources or do we have the skillset to actually win against Sora? And at the end of it, people figure out that We, in fact, did and we put out a model out there that was better than Sora in a few months.
Anastasis GermanidisGUEST
57:59
And I remember we released it at CVPR, which is one of the big computer vision conferences.
Anastasis GermanidisGUEST
58:05
And it was very gratifying for the team that everyone was talking about Gen 3 and the model that we'd released.
Anastasis GermanidisGUEST
58:13
So I think it was one of those formative moments of the company where people realized that...

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.