Torsten Sattler

We have extended the deadline for paper submissions to July 1st.

This is consistent with what we observed when we did the same experiment for ECCV 2024.

Attending @cvprconference.bsky.social and looking for a PhD or postdoc position in the area of 3d reconstruction (Gaussian splatting, nerfs, scene understanding, etc.)? Find me or drop me an email ;)

Come to our poster and learn about our working privacy-preserving localization.

Yes, this is not yet decided and will, among others, depend on ICCV's policy on streaming / recording workshops and the technical equipment we will have.

The workshop is organized by Zuzana Kukelova, Gabriella Flood, Viktor Larsson, Akihiro Sugimoto, and myself.

We accept paper submissions on topics related to calibration and pose estimation, with a deadline on June 28th. See the website for a call for papers.
We also have exciting invited speakers (Richard Hartley, Eric Brachmann, Gabriela Csurka, Fredrik Kahl).

Working on camera calibration or camera pose estimation? Want to finally know how calibrations and poses are obtained for the data you ae using?
Come join us at the CALIPOSE workshop at @iccv.bsky.social

Details are on the workshop website: sites.google.com/view/calipos...

I guess it has to be repeated for each new reviewing round: "See weaknesses" is not a justification of a review rating. And certainly not informative regarding what the authors should focus on in the rebuttal. #CVPR2025

That is to say, I think it makes not much sense to fear what others might do. Do the research that you are interested in and I am sure it will be relevant as you bring a unique perspective to the field.

Regarding progress in a field: I feel that a large part of the community is nomadic in the sense that they move from one area to another. E.g., there was a big rush on NeRFs, which has largely died down. Many seem to move on once the low-hanging fruits have been taken.

I think you are selling yourself very short here. Don't think you would be irrelevant anywhere. Being in a place where you are happy and comfortable is very important. Also, Turkey seems to produce great researchers, so it seems like a good place to be.

Maybe an issue of having enough training data? GANs can work reasonably well with rather limited data.

2025 R&D Software Engineering Intern (PhD, Publishing) – Careers – Niantic Labs

Niantic is looking for research interns for next summer in London. Interested in the next generation of reconstruction, mapping and visual relocalization? Apply! (And send me a DM for good measure). Let's explore what the bitter-sweet lesson can do for 3D vision.

nianticlabs.com/careers/open...

Indeed, the original LO RANSAC paper (DAGM 2003) proposes multiple LO variants.

... are nicely distributed over the images, you will likely get a better model than if all of them are in a small area in the images. In the latter case, the same level of correspondences noise has a much higher impact on model accuracy.

... correspondences. Still, your model will have uncertainties, which (among potentially other things) depend on the inlier noise and the spatial configuration of the correspondences used to compute it. E.g., think of estimating the essential matrix from 20 correspondences. If these correspondences

There multiple things here: a minimal solver will return an estimate that has 0 error on the correspondences in the minimal sample. Thus, the model will be noisy / have uncertainty. LO helps improve things by fitting the model to multiple data points, thus better accounting for noise in the ...

... correspondences used to estimate the model) and the positions of the correspondences. See for examples this paper on handling uncertainty rahulraguram.com/assets/pdf/R... (a very good general treatment of uncertainties can be found in Wolfgang Fõrstner's book).

Assuming that you are fitting the "right" model (a model that can explain what you are observing), the optimal threshold not only depends the noise in the correspondences, but also on the uncertainty in the estimated model (which is a function of the inlier noise but also of the actual ...

By the time that these decisions are made (you probably want to wait until you have the camera ready version), I'd assume that author identities are anyways known.

Unfortunately, a too large part of the reviewer pool ignores almost all emails. Probably the only way to get their attention is to threaten with desk-rejects of their submissions, which seems a bit out of proportion.

Indeed. CV conferences traditionally ask ACs to submit reviewer preferences. Certainly more work, especially since OpenReview and TPMS scores are not always reliable. Probably less chances for collusions than with reviewer bidding though.

The common approach nowadays is that the program chairs make a shortlist of candidates (typically based on AC suggestions) and then select an independent committee (researchers without a conflict of interest with any of the papers) that decides the awards

Ask a PhD student to search for the title on Google / Google Scholar / OpenReview and the authors' websites 😉

... getting good reviewers often means getting a good AC who is familiar with the field and know the right people to review the paper. Getting a good AC often means the PCs doing a good job with assigning papers to ACs. All parts of the chain are important.

... decisions) can certainly be made easier (e.g., don't require meta reviews for clear cases), but can't be fully automated. I don't think we can automate anything close to 90%.

I agree that getting good reviewers (knowledgeable about the field, willing to put in work) is important. But ...

I tend to (partially) disagree. The only part that could currently be automated is sending reminders (in my experience, people are more likely to respond to personalized requests). The other parts (assigning reviewers, leading discussions, checking reviews & rebuttals, justifying and checking ...