jvgemert.bsky.social
Head of the Computer Vision lab; TU Delft.
- Fundamental empirical Deep Learning research
- Visual inductive priors for data efficiency
Web: https://jvgemert.github.io/
71 posts
1,222 followers
203 following
Regular Contributor
Active Commenter
comment in response to
post
Did you have enough capsules for everyone? 😉
comment in response to
post
That's a great way to draw an audience!
Helium included? 😉.
comment in response to
post
What comes after foundation/frontier model collapse? The long term view: Seldon models 😉
comment in response to
post
Empirical vs Armchair.
Do you think it's any different for the ancient philosophers?
Much selection bias: only the Armchairs that accidentally guessed it correctly get credits when empirical validated post-hoc.
comment in response to
post
en fait?
comment in response to
post
It feels half-finished when you end positive 😉
(I'm kidding; I appreciate this a lot, the younger crowd often think academia is a sterile cathedral, omnipotently distilling instant Truth)
comment in response to
post
Too cheesy for your tastes?
comment in response to
post
I'm a fan too! ☺️
comment in response to
post
I also liked it!
You may also enjoy Speed of Dark: en.m.wikipedia.org/wiki/Speed_o...
comment in response to
post
They're breaking my heart! 💔
comment in response to
post
As an AC, i greatly appreciate it if borderline reviewers take my POV and make a forced binary in/out decision and motivate it.
The rating can stay borderline, but a motivated forced decision greatly helps me understand and consider the reasoning, and lead the discussion.
comment in response to
post
I got 22 as an AC, which I find a bit excessive 😅.
comment in response to
post
Wonderful initiative!
comment in response to
post
Thank you! How would such tuning for retrieval work? Sharing INRs across images? Sounds interesting! 🙂
comment in response to
post
There's a paper about that :).
The importance of stupidity in scientific research
journals.biologists.com/jcs/article/...
comment in response to
post
ICCV is one of the major computer vision conferences (multiple thousands of attendants).
This OR glitch could have been a happy little accident: You would definitely draw a crowd at ICCV! (I'm a fan of your work 🤩).
comment in response to
post
Great work! Reducing AI data is important 🙂.
I was wondering, on what data is the used LLaVA captioner trained on?
comment in response to
post
It's the journey that's to be enjoyed, the peak is just an excuse 🙂
comment in response to
post
The Corleones of AI (?)
comment in response to
post
Why not construct a fully controlled dataset yourself?
(No free lunch).
comment in response to
post
Yes, me too..
comment in response to
post
Interesting! What would be a fully controlled (synthetic?) setting to test your vs @davidpicard.bsky.social hypothesis?
comment in response to
post
Wonderful! Can you post a link to the arxiv landing page instead of the pdf link?
Landing pages are much easier to store, share, and include 😊
comment in response to
post
I wish reviewers were honest about their reasons; then I can try to remedy it.
comment in response to
post
I speculate that reviewers write "lack of novelty" when they really mean: "not interesting".