🚨 Preprint alert 🚨
We develop FireANTs, the first Adaptive Riemannian Optimization for Multi-Scale Diffeomorphic Registration.
This represents a new state-of-the-art registration method 👇
paper:
code:
1/n
#medicalimaging
Excited that our paper has been accepted at
#CVPR2023
with three 5/5 reviews.
So what did we do? New SOTA in some task? Quite the opposite, actually. Stay tuned! 😉
(P.S. Also it's my first
@CVPR
paper so I'm hella excited!!)
I'm very excited to announce our
#CVPR2023
✨highlight✨ paper "Beyond mAP: Towards better evaluation of instance segmentation"
Project page:
Arxiv:
Code:
A 🧵
Excited that our paper has been accepted at
#CVPR2023
with three 5/5 reviews.
So what did we do? New SOTA in some task? Quite the opposite, actually. Stay tuned! 😉
(P.S. Also it's my first
@CVPR
paper so I'm hella excited!!)
@BeerBicepsGuy
cant believe hearing this from a productivity youtuber.
for me, the apple ecosystem provides great convenience when context switching (music, papers, books, etc) from one device to another (heck, even the airpods switch automatically when you switch to a different apple device)
End-of-week update: I'll be joining Amazon as an Applied Scientist Intern this summer in Sunnyvale, CA ✨
Excited to finally have some industrial research experience!
@BeerBicepsGuy
even the clipboard is shared, providing a smooth flow when working on my mac and needing to copy some important screenshot/note to iPad when taking notes.
you're simply using the iPhone in isolation, which is useless. the apple ecosystem makes you super productive
@MilesCranmer
there is technically some "cheating" because the paper uses a floating-point number with arbitrary precision.
Since the amount of bits required is very low(!) this paper is a good candidate for compressed representations maybe. But its definitely not a "single" parameter.
I don't think Musk is interested in moderating/censoring/etc any academic tweets, so leaving academic twitter seems to be a bit of an overreaction. thoughts?
I sincerely hope people in my communities don't leave Twitter. I think there's a lot of value in it right now, and it's unlikely to be reproduced on another platform. And I fear people preemptively moving may permanently splinter the community.
If you're attending MICCAI this year, be sure to check out our poster titled "Self-Supervised Vessel Enhancement Using Flow-Based Consistencies" on Sep 28 (Tuesday) from 14:00-15:30 EDT. 🤩
Paper:
Code:
@MICCAI_Society
#MICCAI2021
Great choice of venue!
#CVPR2023
I'm presenting our work on evaluating instance segmentation Wed, and I'm looking forward to meeting new people. Come say hi (or DM) 😄
@Ethan_smith_20
this is by design (all datasets are projections of reality), the important detail is to see how much of Z is “realizable” with scaling and data
should we have different tracks for conferences with high/low computational requirements?
Like a track for papers using < 48GB GPU compute, and another track for others 🤔😛
rebuttal.pdf uploaded ✅
rebuttal.mp3 uploaded ✅
Whats more fun than writing a rebuttal? Making an AI-generated diss track for the
#ICCV2023
reviewers (ft.
@atemyipod
)
it's amazing how effortless this is to make
#iccv
loss functions only tell you "what" to achieve, not "how" to achieve
this sounds trivial, but the revelations I've had in the past couple of weeks has been astonishing!
@CSProfKGD
this is a very personal take, but I think the student is starting off with the wrong incentives/goals (publish something in the next big conference) rather than finding cool and impactful problems and solving them
AI influencer clown literally just googled "hardest CS/math problems" and equated them to an admission at MIT/Stanford 🤡
if you prove P=NP, MIT/Stanford would be the last place u would be going to
Unironically think we can solve most of our problems by just assigning elite university admissions bounties
-> solve Riemann hyphothesis -> guaranteed admission and full ride to MIT for you and 6 others of your choice
-> p=NP -> Stanford for 10
Etc.
Utilize talent of
its scary how many PhD students, who are absolutely horrible mentors, want to become professors.
Mentorship training should be highly encouraged/mandated in PhD programs imo.
new reviewers with zero math knowledge and unwillingness to catch up with the background have absolutely ruined the review quality
there should be an exam for screening out such people
Embarrassing
#NeurIPS
review:
- Two reviewers asked what is evidence lower bound
- One reviewer asked comparison with a CVPR'23 paper (the conference held after submission...)
- One reviewer said that analyzing existing algorithms w./ no new algorithm is "below NeurIPS bar"
Me: Hey, here's a method that runs 20x faster and uses 8x less memory than SOTA method [A], and performs at par with [A], giving us a good approximation under a compute budget.
one of the
#ICCV2023
papers I'm reviewing has another reviewer saying the following
"The results are good but I'm not surprised..." without any justification
pls don't do this. hindsight is always 20/20
a few days ago I discovered (or learned) that for positive numbers x and n, if n is a power of 2, x%n is the same as x & (n-1)
My appreciation for tricks like these has increased exponentially since doing low level CUDA programming, and thinking in terms of cpu cycles
@NathanielEDavid
shouldn’t the y axis compute mean mutations per lifespan instead? otherwise, animals with lower lifespan reproduce more per year and therefore have more mutations per year …
I'm very excited to announce our
#CVPR2023
✨highlight✨ paper "Beyond mAP: Towards better evaluation of instance segmentation"
Project page:
Arxiv:
Code:
A 🧵
@debarghya_das
I was AIR 57 a while ago, and I know what you’re talking about. But even the precision of the exam is low (I had a fair share of classmates who were spoonfed everything by their coaching, and struggled exponentially more with the undergrad curriculum on their own)
reminds me of a dialogue from 3 idiots
"school jane ke liye fees thodi lagte h, uniform lagta h, uniform!"
"you dont need fees to attend school, all you need is a uniform"
We present SplatArmor, a fully articulated Gaussian splatting model for human avatars. Our model includes both rigid and non-rigid skinning components, and a Neural Color Field for implicit color regularization.
Project page:
I'm delighted to share that our paper on combining behavior cloning and GAIL for on-policy sample-efficient imitation learning has been accepted at CoRL 2020.
Arxiv:
@vishalkhatri22
@BeerBicepsGuy
thats what i said about using iphone in isolation. using iPhone in isolation is a status move, but using the apple ecosystem as a whole is for power users
@CSProfKGD
The problem with open dialog (what I've heard from my friends who regularly submit to ICML,NeurIPS,etc) is that it leads to reviewer fatigue, and they end up not responding at all.
I think a single rebuttal round is fine, as long as there is a per-reviewer character/page limit.
Just when I thought I was done with the
#ICCV2023
submission, I realized I have
#MICCAI2023
submissions, supplementary materials,
#CVPR2023
camera-ready & code release, visa stuff, and then maybe I can go to sleep.
am i doing academia right?
@BeerBicepsGuy
I've used Windows, Linux (3 variants) and MacOS. Cannot complain when Linux has trash presentation tools, and Mac has the best of both worlds - awesome presentation tools (good for preparing talks,etc) and programming environment of Linux.
From "You and Your Research" by Richard Hamming
"Great scientists tolerate ambiguity very well. They believe the theory enough to go ahead; they doubt it enough to notice the errors and faults ..."
"Great contributions are rarely done by adding another decimal place."
I just asked GPT4 to write my thesis, and then used the “expert thesis committee” evaluation mode to rate the thesis. Glad to inform that I’ve passed my thesis defence👨🎓
I'm very excited to announce our
#CVPR2023
✨highlight✨ paper "Beyond mAP: Towards better evaluation of instance segmentation"
Project page:
Arxiv:
Code:
A 🧵
what are some good medical imaging datasets or tasks (not registration) that require long-range image feature dependencies?
For e.g. in segmentation datasets, most baselines seem to do well with only small local patches or axial image slices as input
#medicalimaging
this is a bit misleading…. how is a major flaw in the paper (GPT evaluating itself) addressed by collection of data with or without consent (which is another issue btw)
FINAL UPDATE: On June 24th, Armando Solar-Lezama (Professor in EECS and COO/Associate Director of CSAIL, MIT), Tonio Buonassisi (Professor of Mechanical Engineering, MIT), and Yoon Kim (Assistant Professor in EECS and CSAIL, MIT) released a public statement regarding the paper.
17-year-old Grandmaster D Gukesh is the Champion of FIDE Candidates Chess Tournament 2024!!
Gukesh scored 9/14 points to clinch the first place. A huge moment for Indian Chess! After Vishy Anand, he is only the second Indian player to win the Candidates Tournament.
With this
"The method uses mesh which has been used before" - reject
by that line of reasoning all papers after the original nerf paper should be rejected because nerf is not novel anymore
@SenguptRoni
It all starts with preparation. I highly recommend storyboarding lectures/talks in analog format first. I use sticky notes to first brainstorm and then organize them into a story.
The biggest mistake I find most people do is rush to their presenter software, that comes LATER.
@Michael_J_Black
i think it’s hurting people in the sense that people cannot discuss their paper for months because of the ban, thus hurting the visibility of small labs
We are delighted to share that our oral workshop paper - "MA3: Model Agnostic Adversarial Augmentation for Few Shot learning" won the Best Paper Award in CVPR Visual Learning with Limited Labels workshop 2020. Joint work with
@shirsendush
Oral talk:
#PrincetonU
will increase graduate fellowship and stipend rates by an average of 25% to about $40,000 for doctoral candidates during the 10-month academic year.
It is the University’s largest one-year increase in graduate student stipend rates.
@miniapeur
the excellent geometric proof of "sum of inverse squares of natural numbers is equal to pi^2/6" illustrated by
@3blue1brown
I never thought myself about where does pi come from in the infinite sum, or why is it squared until I watched the proof
@ReheSamay
has been revolutionizing the chess scene in India during the lockdown, so what could be a better new year's gift than a chess bot of the Blunder Master himself.
Also features bots on
@sagarchess1
and
@viditchess
with respective difficulty levels 1/n
@AlphaSignalAI
what I read - its a new MLLM that has fit the given data distribution than other models because of better data/training/etc. There is no inherent reasoning capability
@naval
dont think it’s an accurate description, as per debroglie’s rule, small particles have more “wave like” characteristics and they show their particle behavior only when measured, but it doesn’t mean some particles are “teleported” to other universes
@jbhuang0604
on a related note, is it ok to cite the reviewer guidelines and slides shared to call out bad reviewer behavior
"not novel enough for CVPR -> reject"
"not novel as its done before" (provides no citations to back up claim)
@ccanonne_
@yoavgo
why is it an insult if OP have valid reasons for it (gpt evaluating itself is unacceptable)
i might as well start training on the test data and show 100% accuracy on all vision benchmarks
@sethrosen
@chrisalbon
yeah, but once you want to debug some complex code that ChatGPT wrote but dont understand it, things become problematic.
and before u say "I'll ask chatgpt to debug it" its going to be difficult bc u most likely don't understand the nature of the bug, so ur prompts will be bad