Inspired by
@dimadamen
and
@CSProfKGD
, here is a list of my rejections. Hope it helps younger researchers that while rejections are certainly annoying, they happen to everyone. Hope it will help someone feel better about themselves if they are down. 1/N
Seriously impressed by what modern local features can do. This is certainly a hard image pair, even for humans. I am very surprised by how well modern local features work here. 1/n
Happy to announce that we got one paper accepted to
#CVPR2022
and looking forward to present it at the conference. We also had 4 papers rejected. For these, we received valuable feedback that will help us improve these works.
Not sure whether this helps, but I find it easier to inspect a dataset of images by looking at a video than individual images. Here is a simple script to generate a video from png images: ffmpeg -framerate 1 -pattern_type glob -i '*.png' -c:v libx264 video.mp4
If you are attending
@ICCVConference
and are interest in 3D vision, please join us at our tutorial on camera geometry estimation (). The tutorial will start today (not tomorrow as in the printed program) at 13:30 in room W07.
Call for papers now open!
Registration deadline: May 26 (after the release of ECCV reviews but before final decisions)
Submission deadline: June 2
Supplementary deadline: June 9
I am looking to hire PhD students. If you are interested, don't miss to apply through the
#ELLISPhD
program to benefit from co-supervision by an internationally acclaimed researcher and other networking opportunities.
I am looking for a local feature detector and descriptor that runs at 5ms or less on small images (640px or smaller), ideally competitive to SIFT. Any suggestions would be very welcome. The fastest I currently have is .
My group has two open PhD positions on 3D mapping with privacy control respectively building 3D maps that model the fact that our world is dynamic. If you are interested, apply through ELLIS or directly to me per email.
Shameless self-promotion: Happy to share that our (work with
@f2011228
, Fredrik Kahl) paper on recovering point clouds from line clouds in the context of privacy-preserving localization was accepted to
#CVPR2021
. Paper available on arXiv:
Interested in
#visuallocalization
and
#placerecognition
? Check out our
#iccv2021
workshop on long-term visual localization and our challenges: The winner and runner-up of each challenge receives prize money. Challenge deadline: October 1st
Thank you very much to everyone who organized a workshop at
@CVPR
! I very much enjoyed the workshops this year and appreciate all the hard work that went into organizing them.
It's amazing to see how much stronger Master students are nowadays compared to when I finished my studies. I quite often see extensive course work on computer vision and machine learning, including cool projects. 1/n
The Universal Image Embedding Challenge () of our
@eccvconf
Instance-Level Recognition workshop () is online now! The workshop is co-organized by
@0xtob
and
@giotolias
among others.
@MattNiessner
I think requiring research experience before starting the PhD is on the long term is not healthy: at some point, students have to find out who they are and what they want to do in life. I think doing that before making a long-term commitment (starting a PhD) would be good. 1/n
You are working on
#3DComputerVisiom
and are dissatisfied with your ECCV reviews? Consider submitting your paper to
@3DVconf
. Paper registration deadline is May 26, paper submission deadline is June 2nd. Join us in beautiful Prag in September, either in person or virtually.
This Wednesday,
@AjdDavison
will be giving a talk on "Representations and Computational Patterns in Spatial AI" in the
#3dgv
seminar series.
@StefanLeuteneg1
will join as a panelist. Join via on March 31st at 11:00 UK time.
Shameless self-promotion, part 2: we will present out
@eccvconf
paper "ViewFormer: NeRF-free Neural Rendering from Few Images Using Transformers" (with
@jonaskulhanek
, Erik Derner, Robert Babuska) in the morning poster session on Thursday, Oct. 27.
Are you working on traditional computer vision techniques or in the intersection of learning and traditional vision? Your paper was not ready and ECCV is too far away? Maybe our IJCV special issue (deadline Jan. 31st) is of interest to you
However, I am also a bit worried whether we start overemphasizing publications as part of the MSc studies. Are we starting to loose promising students because they have less opportunities to publish, e.g., because they come from countries with less strong vision communities?
Old school computer vision paper with Yukai Lin, Viktor Larsson, Marcel Geppert, Zuzana Kukelova and
@mapo1
at
#ECCV2020
: Infrastructure-based multi-camera calibration using radial projections
1/n
Very cool project presented today at
#3DV2020
: Kyle Wilson, Scott Wehrwein, Visualizing Spectral Bundle Adjustment Uncertainty,
They have very interesting visualization
@fdellaert
I think you would like this.
The first talk in our virtual seminar on 3D vision + geometry processing by
@mapo1
starts in about 30 mins (12:00 (noon) central European time). Watch it live at . For more details, see .
#springerdejavu
: ECCV authors: here is a carefully prepared paper, we have spend a lot of time placing figures and optimizing tables for readability and space. Springer: here is your final version with randomly placed figures and completely reformatted tables 😂😂😂
I really like the poster sessions at
#3dv2020
. 3dv uses spatialchat to create poster rooms, where posters are shown. You can navigate in 2D and you only hear people close to you. It nicely reassembles real poster sessions, including giving you an idea which posters are in demand.
The Unreasonable Effectiveness of Pre-Trained Features for Camera Pose Refinement
Gabriele Trivigno,
@masone_carlo
,
@bcaputo_iit
,
@SattlerTorsten
tl;dr: perceptual metric+particle-filter-based optimizer->pose hypothesis->render&compar->pose discrepancy
DAGM GCPR, VMV, and VCBM are starting tomorrow and are free to attend virtually. We have exciting keynote speakers, invited talks, papers, and tutorials. For more information, see
We have exciting keynotes, papers, tutorials, and demos at
#3dv2022
. Please consider joining us, either in person in beautiful Prague, or virtually.
@3DVconf
takes place from Sept. 12th to 15th.
D-20 until
#3dv2022
Prague is full of amazing spots to visit! Here is a shortlist:
Let us know if we missed your favorite spot and we will add it ;)
Register here:
If you are attending
@CVPR
, either in person or virtually, don't forget to vote. The deadline is Thursday, 6pm central time. These are the motions that would impact future CVPRs and ICCVs: 1/n
Do bad reviewers complain about bad reviews? One of the things that puzzles me about reviews is that everyone (me definitely included) complains about their quality, but few of us seem to be motivated to write better ones themselves.
@ducha_aiki
We are also ranked 14th in the world and 2nd in Europe in computer vision according to ( and ). In general, Prague has a very strong computer vision community.
Our paper on "Large-scale, real-time visual-inertial localization" (Simon Lynen, Bernhard Zeisl, Dror Aiger, Michael Bosse, Joel Hesch,
@mapo1
,
@rsiegwart
, Torsten Sattler) got published in IJRR:
Also available on arxiv:
Reviewing for
#iccv2021
and late with your reviews? Please contact the area chair / meta reviewer for your paper (through cmt) and let them know when you expect the review to be finished. It tremendously helps them to understand which papers need emergency reviewers.
As part of our Instance Level-Recognition Workshop () at
#iccv2021
, we are organizing challenges on landmark recognition () and landmark retrieval ().
Shameless self-advertisement part III: please join us tomorrow for our
#ECCV2020
workshop on long-term localization: The workshop takes place 10:00-12:00 and 18:00-20:00 UTC+1.
#ICCV2021
AC musings: I often see multiple points with one sentence each to list the strengths of a paper. It helps everyone (ACs, authors, reviewers) a lot to also state why these strengths are important, e.g., why is new dataset interesting? Don't be afraid to be positive :)
#iccv2021
AC report: 7 days to go (including today), 29 out of 83 reviews received (+8 reviews since 2 days ago), only 3 papers without any reviews. Big thanks to the reviewers who replied to my mails and added details to their reviews!
More shameless self-advertisement. We are presenting our paper Erik Stenborg, Torsten Sattler, lars Hammarstrand, Using Image Sequences for Long-Term Visual Localization, today at
#3dv2020
. If you are attending 3DV, please consider stopping by. Paper and code coming soon.
Together with
@hulahuub
@mapo1
and others, I will be organizing an ECCV 2020 workshop on long-term visual localization:
We offer three challenges:
Submission deadline: August 15th
Interested in NeRFs, but bored by vowel grids and point clouds? Come visit our
@ICCVConference
poster. Poster 82 kn the south room, starting 10:30. Project page:
More shameless self-advertisement: If you are at
#3DV2020
today, please consider stopping by at our poster: Anders Sunegård, Lennart Svensson, Torsten Sattler, Deep LIDAR localization using optical flow sensor-map correspondences. Paper coming soon online.
As PhD student: did not even submit between ICCV 2009 and ICCV 2011 because nothing that I tried was working. Considered quiting my PhD if ICCV 2011 paper was not accepted. Papers rejected from CVPR 2012, CVPR 2013, CVPR 2014. 2 / N
If you are attending
#3DV2020
, please stop by our poster today:
@Poyonoz
, Martin Humenberger, Gabriela Csurka, Yohann Cabon, Torsten Sattler, Benchmarking Image Retrieval for Visual Localization, 3DV 2020,
Times (CET): 7am and 5:30pm
@giffmana
In general: Use a modern RANSAC implementation that uses the T_d,d or SPRT tests (making things faster) and local optimization (improving model quality and reducing # iterations), e.g., MAGSAC++. If you have confidences, you can try PROSAC-style sampling or the NAPSAC-style ...
The 2nd talk in our virtual seminar on 3D vision and geometry starts in about 1h 40mins (11am west coast time). Katerina Fragkiadaki will talk about 3D Vision with 3D View-Predictive Neural Scene representations:
Kind reminder: The
@eccvconf
paper submission deadline is on March 7th at 22:00 (not midnight) CET (not PST). provides a countdown to avoid issues with time zone conversions.
#eccv2024
#eccv
This is an amazing dataset showing the triumph of COLMAP "COLMAP, a mature photogrammetry framework, provides 3D annotations that are treated as ground truth"
@mapo1
@SattlerTorsten
@MattNiessner
So in the long term, you will likely not increase the quality of the applicants. However, it will put further stress on students (and a PhD is already stressful enough). For reference, I never did internships or knew about computer vision before starting my PhD. 6/6
Quick summary of our activities at
#ICCV2021
@ICCV_2021
:
In session 5 (Tuesday 4pm to 5pm EDT and Thursday 9am to 10am EDT), we present two papers: 1/n
Shameless self-promotion: we will present our paper Objects can move: 3D change detection by geometric transformation consistency () (
@KaterinaAdam5
,
@SattlerTorsten
, Konstantinos Karantzalos, Tomas Pajdla) on Tuesday, October 25th, morning. Poster 104.
@docmilanfar
Very sorry to hear about this. We had tried to catch such cases and tried to get reviewers to improve their reviews (or get an emergency reviewer). Unfortunately, we did not manage for all cases (but at more than 25k reviews, this is hard to do).
Also starting in about 30 minutes: tutorial on large-scale visual localization, organized by
@giotolias
,
@1avr1e
, Zuzana Kukelova, Sudipta Sinha,
@mapo1
,
@eric_brachmann
and myself. Join via
@CSProfKGD
Indeed it is. We had 4 papers rejected, but got constructive criticism back that will help improve the paper. Nevertheless, I am very proud of these papers and happy to work with such great people.
@AljosaOsep
I did 4 emergency reviews this time around (10 reviews in total). To be honest, I am starting to regret investing a lot of my time looking at some of the reviews that I am getting.
@Michael_J_Black
Very nice post! Should become part of the reviewer guidelines. It reminds me of one of Konrad Schindler's talks, where he advocates to measure a paper based on whether it will be of interest to the community.
What if you actually could build Rome in a day? Prof Richard Hartley is one of the founders of Multiview Geometry, which establishes the construction of 3D models from sets of images or videos. He receives the Hannan Medal for his pioneering work.
LAST CALL:
#ECCV2024
has a “paper registration” deadline. Deadlines are strict. NO registrations & NO author list changes will be accepted after the deadline! NO registration, NO submission! NO exceptions!
Deadline: Feb 29, 2024 01:00 PM PST
Dates:
@jacobsn
@CVPR
I think this could be extended to also account for environmental impact of a paper. We could have some form of "green" best paper award and could try to look at the tradeoff between energy consumption and improvement of a method.
I would definitely not hire myself as a PhD student these days :)
I think it is a testament to how the field has grown to see so many students finishing their Master with such strong backgrounds. 3/n
@MattNiessner
(born in the "wrong" country, not enough money to afford moving to another country or go to university in the first place). Those fortune enough to have access to good education need to compete even earlier to go to a good university in order to have a (long) shot at 4/n
@docmilanfar
We will have this discussion soon. So far, there was too much to do for that. It will need to be coordinated between the conferences. In my opinion, it should at least be sharing lists of unreliable reviewers and ACs, potentially submission bans for 1-2 years (similar to a ...
One thought about review forms: why not remove the accept / borderline / reject choice and only ask reviewers to summarize and list strength and weaknesses? If you want to accept / reject a paper, you would be forced to argue for it. 1/n
Shameless self-promotion, part 3: we will present our
@eccvconf
paper "MeshLoc: Mesh-Based Visual Localization" (with
@VojtechPanek
and Zuzana Kukelova) in the poster session tomorrow morning (Thursday, October 27):
#ICCV2021
AC report (inspired by
@CSProfKGD
): 3 days to go, 41 out of 83 reviews received (49.4%), 2 papers with 0 reviews, 8 papers with 1 review each, 9 papers with 2 reviews each, 5 papers with 3 reviews each, 0 papers with 4 reviews.
The submission website is now open!
Remember to register your paper at by May 26th
COUNTDOWN to
#3dv2022
submission deadlines ()⏳
Paper Registration: 8 days left
Paper Submission: 15 days left
Supplementary Submission: 22 days left
If you haven't finished your
@eccvconf
reviews yet, please reach out to the area chairs handling your papers to let them know when you will be submitting them.
I can only recommend attending. One of my favorite workshops over the last few years. If you can't attend the full workshop, I recommend to at least attend the panel discussion.
We are pleased to welcome
@dimadamen
,
@FidlerSanja
and
@zeynepakata
as our guest speakers at
#WiCV
. They will be joined by Rahul Sukthankar for a panel discussing diversity in computer vision. Leave your anonymous questions for the panel here:
#ICCV2021
AC report: about 8h to go, 65 reviews received and 5 more promised out of 83 reviews (84.3%). 2/0/6/11/5 papers with 0/1/2/3/4 reviews. Thanks to all reviewers for their great work.
@MattNiessner
At some point, there should be enough time to develop as a person. Likely, this will not happen during a PhD. A PhD is a marathon, not a sprint. If we require students to do research beforehand, we ask them to run a long-distance race before the marathon. 2/n
@MattNiessner
How many students will we burn out by asking for too much in too little time? In order to do research that will lead to publications, you need to have access to good mentors (and thus good universities). That will prevent a large part of the population to begin with 3/n
@taiyasaki
I think writing a rebuttal is an important way to get feedback: are my arguments convincing? What can I improve? Compared to the initial review, commenting on the rebuttal is fast. I think that round of feedback helps improve papers, and isn't this a main function of reviews?