Hacker News Reader

Hacker News

Comment by 1letterunixname | Hacker News Reader

Full thread

1letterunixname

One could also write an image crawler to execute from AWS.

The key is finding quality source material.

It may become necessary to:

1. Fly drones around to gather data in the world at present time

2. Direct humans to take high-quality photos of particular subjects. Fiverr for photography

View on HN

I've built a site to pay people for images and annotations. [1] I'm trying to onboard my first paid users right now. The plan is to build out a high quality 50k image license plate recognition dataset as a proof of concept.

Right now we own all of the datasets on the site and the idea is to license them out to companies while making them available to researchers under a non-commercial license. The market might take it a different direction to be more of a marketplace or Github style hosting. Email in bio if anyone wants to chat about this.

Also, if anyone wants to get paid 10 cents an image to take pictures of North American license plates, get in touch. Need about 1000 from each state. It's probably below most people's pay grade on here but there is a whole reverse bidding system, so you can always bid higher than 10 cents. Some user studies with a shared screen would be super helpful as well.

1. https://mekabytes.com

Wow, interesting project. Not sure how many people you can entice into providing 10k photos of license plates, firearms, and children in pools. I kid, but are you building an alert system for a superhero?

Ha! It is a bit public safety focused right now. The first two I will really build out (read: pay people to help with) are license plate recognition and vehicle make/model/year identification. I think those have a decent market.

The firearms one is tricky and I'm not sure it will ever be licensed commercially since we don't own the footage that the images are taken from. Valuable dataset in terms of what it could provide though, like an early warning system for active shooters.

Some day I'll have enough free time and just make a billiards dataset to help me find shots.

Have a good look at privacy laws in the states/countries of your users, in some things like this may be against the law or subject to restrictions. See:

https://www.ncsl.org/technology-and-communication/automated-...

Thank you! That's honestly very helpful.

You're welcome & much good luck with your project!

Unattended security gates at a local campground are being used to gather data. To get access to the campground, you have to provide your plate number, vehicle make & model, and vehicle color. The gate only checks for the plate number, but the images captured can be used to build out a vehicle make/model ML model.

Maybe partner with such a system to gather your data?

I have several thousand images of vehicles on the roadway and parking lots that I used to understand the backend ML of an ALPR system. Selecting cameras was the most difficult part of the project. I did not attempt to determine the make/model. My patience for labeling had worn too thin.

Overall, as invasive as ALPR system are, the cat has long left the bag. I doubt the cat will ever return.

Hey there, just checked out the site and signed up. I’m not seeing anything about how to get paid for uploads. Can you provide some direction here? Thanks!

It is a little hidden still. On your profile page there should be a sign up link that will take you to Stripe. Feel free to email me. I will post two solicitations tomorrow, one for annotation work and one for plate photographs.

It is restricted to US only right now (Stripe setting) but I will change that if there is non-US interest and the country is supported.

I've only run through the whole process once so hopefully there are no hiccups. Thank you!

What is the cost for each dataset for commercial uses?

We haven't figured out commercial pricing yet. It may be a scale based on company revenue. Targeting bigger businesses though, it will be less than it would cost to develop the model independently, but likely trying to target low 5 figures.

The datasets are also available and a non-commercial Creative Commons license. You can pay $5 for a download and rehost it elsewhere, or just wait for me to upload it to Kaggle or Hugging Face

"Also, if anyone wants to get paid 10 cents an image to take pictures of North American license plates, get in touch."

Your blatant disregard for privacy is shocking, but perhaps unsurprising in the field. I guess you also didn't through the enormous risks for the photographer.

This will be the new captcha challenge—“take a picture of 14 crosswalks”

It will send you a link to an app that only runs on an iPhone so it can verify you actually took the pics from the phone.

That would not be a11y friendly!

Why exactly, they should know how to use crosswalks by that age.

a11y means accessibility. Apparently this is a common enough jargon in that part of the industry.

Is the jargon going too far?

I the j4n going too far? English always had a dislike for long words, so turning long and s11d words into shorter forms is the next logical step in l6e evolution. Certainly beats having to spell w12e sauce in its entirety /s

At least w12e is easier to pronounce:-)

At some point LLMs will have a easier time understanding English than actual human beings.

Expecting people to know this is going too far. I mean usually the acronym gives a hint, this is just straight up medieval druid vibes. "Oh the field which begins with a followed by our savior's pious 11 letters and then a y"

Neither are current captchas: they're ina11 to non-US residents (/former residents).

(We don't have 'crosswalks'. 'Fire hydrants' don't look like that. 'Parking meters' don't look like that (or any one standardised thing really). Traffic lights don't look like that and how much of them am I meant to click anyway. School buses don't look like that. Trucks don't look like that. I'm sure there are more.)

And often it's even more subtle. Apparently Americans don't consider pedestrian lights part of the traffic lights. And apparently they only mark the actual light assembly, not the structure holding them up.

Lot's of things where you might think you know what you're doing, but are actually doing it wrong because you lack cultural context..

That doesn’t seem like cultural context at all. That just seems like the arbitrary decision of the person who created the captchas. I am, presumably, from the same culture as them and would have agreed that pole was part of the traffic light.

I know most of HN lives downtown but not everyone is near a crosswalk. And then there are blind people, etc.

I assume non-downtown US residents are still significantly more aware of the term, what it is, and what it looks like.

I'd heard it in films before captcha, but had to guess it's the zebra crossing looking (the meaning is different though, I understand) things. Now it's just something I'm familiar with because of captcha, even more so 'parking meters'. Which is just weird, why do I have to learn something mundane about a specific country in order to prove I'm human and allowed to use a global website?

takes pictures of a screen with the phone

You mean like what google did with street view and convincing everyone to upload their photos on to maps?

Like that but properly organised and with higher quality imagery

I've been wanting this for quite sometime

Or like Mapillary (bought by Facebook, but with a liberal license still)

Reply on news.ycombinator.com