ericbrachmann.bsky.social
Staff Scientist at Niantic
Throws machine learning at traditional computer vision pipelines to see what sticks. Differentiates the non-differentiable.
http://ebrach.github.io
146 posts
1,771 followers
506 following
Regular Contributor
Active Commenter
comment in response to
post
Maybe be an apprentice shoe maker in between.
comment in response to
post
Thought as much. So I guess the only scalable way to finance reviewer rewards would be submission fees - and that's not great in terms of equality of opportunity.
comment in response to
post
Dhruv mentioned millions in sponsorship money, so I thought >1 😉
comment in response to
post
And is my napkin math above on any way realistic? Do conferences like CVPR have $10M laying on the side to spend extra on reviewing? @wjscheirer.bsky.social
comment in response to
post
Now I really do wonder what would actually happen with monetary incentives. 🤔 For example: If you pay too little, experts might not be interested. If you pay too much, non-experts might try to game the system to get in and score some cash.
comment in response to
post
Is there enough money at the current scale? Assuming $10M available to spend on reviewing, with 10k submissions that is $1k per paper to pay 4 experts to spend several hours on this.
You were complaining about too many grad student reviewers recently. That's the only group I imagine would bite 😜
comment in response to
post
It has only been on arXiv for a few days... Sheesh, give them a break.
comment in response to
post
Now I wonder whether it's a "grad student" problem. Not sure... People do not necessarily get wiser while climbing the career ladder. 😉
comment in response to
post
Personally, as a reviewer, I tend to share suggestions if I have any. But I keep it modest and subjective. I try to avoid a patronizing tone explaining what the authors should have done. As AC, I do not add suggestions, but sometimes I might add encouragement to sweeten the bitter pill of rejection.
comment in response to
post
Don't know. I'll ask someone who has the technical skills to harness this power to its full extent.
comment in response to
post
Oh SLURM, I miss you so much...
comment in response to
post
Looks promising! Let's see how well it fits my data.
comment in response to
post
And rightly so. The lack of engagement here is also preserving my Sunday peace.
comment in response to
post
That's [0, inf) support, right? I need (-inf, inf) support.
comment in response to
post
I certainly agree. Still, most people did not engage, so it makes no sense to criticize them for leaving Twitter. (Also, not sure the "broader public" ever was on Twitter. It has always been comparatively small.)
comment in response to
post
When they felt that their use of the service was supporting destructive forces, they switched the service.
comment in response to
post
Irrespective of whether I agree or not with that take, I think it does not apply to large parts of the CV/ML community. They were not engaging in political discussion, nor in public science communication. They were using the platform to exchange ideas with peers.
comment in response to
post
I think I might enjoy doing reviews again for a change.
comment in response to
post
Agreed. I wonder, too. And the decision to leave X might be easier for mid- to end-career researchers compared to an early-career researcher.
comment in response to
post
Did you renounce MySpace? That seems to have worked.
comment in response to
post
I miss Twitter. I think it helped me build a brand for myself and ultimately advanced my career. But it's gone. If Bluesky dies, it might just be paper posts on LinkedIn for me.
(Still, I was unable to bring myself to delete my X account yet, it's just inactive...)
comment in response to
post
Volunteering as social media chair on Bluesky? :)
comment in response to
post
Learn from the best.
comment in response to
post
And with everything being grey and brown, we might not need RGB anymore. Golden times for vision.
comment in response to
post
There are encouraging results here for Aachen Day-Night, presumably due to using the strong feature backbone by @parskatt.bsky.social. Season changes might be more difficult, but depends on the environment. In urban spaces, the impact of seasons is not that big - except for city parks maybe ;)
comment in response to
post
something = normalisation factor
comment in response to
post
That's ambiguous. There should be one M for each mode making the model multi-modal. Text and images? LMMM. Text, images and audio? LMMMM.
comment in response to
post
Yes, I think that is the official way. It makes me nervous though and makes me triple check the association of name and ID. 🙂 And there is the chance that the reviewer replies to that comment not knowing that it's invisible to others.
comment in response to
post
Yes, generally I do that. Sometimes I like to send private messages which address the reviewer by name, in the hope it makes them feel more responsible.
comment in response to
post
Looks legit. 7-9 could map to "Br...".
comment in response to
post
For example, tragedy or illness are valid reasons for not submitting. There should be a human in the loop before applying drastic sanctions.
comment in response to
post
Thanks, good to know. I guess with these systems it's sometimes not obvious what actually goes out to recipients. It was not you :) (At least this time 😀)
comment in response to
post
I do not know whether there is such a reputation, and if there was I should know better than to support it.
comment in response to
post
Sorry, I should not joke around with this. It's tough for anyone anywhere. And whatever I believe to have recognised in certain people I should really not generalise.
comment in response to
post
You are right. What ETH students really really urgently need is boosts to their confidence 🙃 Thank goodness for these rankings...
comment in response to
post
I guess I would be more convinced if the ranking revealed anything surprising. But to me, the top places are just what you would expect, and further below, the ranking seems to be arbitrary.
But I guess that is what you are saying: What is obvious to me now, would no be obvious to someone new.