• 0 Posts
  • 21 Comments
Joined 1 year ago
cake
Cake day: August 8th, 2023

help-circle
  • In its email, Google states that it is closing down the program because of the "overall increase in the Android OS security posture and feature hardening efforts." This has led to researchers submitting fewer vulnerabilities than before.

    1. Vulnerabilities are found, which shows that the program is successful and needed.
    2. No vulnerabilities are found, no money will have to leave Google.

    Keeping the program will reap the benefits from both no. 1 and no.2 while closing down the program only enables no.2.

    Not hard to see the priorities here....



  • I don't know how to feel about this.

    On the one hand, it's cool that they pushed old electronics way beyond the known limits, but on the other hand is 120p really an accomplishment?

    Even my old Commodore 64 from 1982 was able to produce around 400p when pushed to the limit (I know progressive wasn't thing on tvs then. I'm simplifying things to not end up on a side quest here). The norm was 200p and exploring how far the electronics could go in that resolution would be far more interesting in my opinion.

    If we're just focusing on framerate, I'm pretty sure it would be possible to reach over the kHz limit with 1p.

    Essentially it would be possible to run 1p led-aray at 1MHz or more...



  • mindlight@lemm.eetoOpen Source@lemmy.mlOpen source handwriting OCR?
    ·
    edit-2
    4 months ago

    How good is good do you say?

    We got a pretty good results with CER at 4% and WER at 15%!

    This was on a limited dataset used to test and train which most likely means that if you introduced an even larger dataset with greater variations in handwriting style for testing the numbers might be even worse.

    Very simplified: A risk of a character wrong every 20th character and a word wrong every 7th word. The SER was around 20%.

    There's an reason why no one has released a good model for western letters yet and why companies pay up to 1€ for capturing data from 10 handwritten pages.

    It will come but OCR isn't as sexy as developing text2image solutions.




  • To train an AI to recognize handwriting you need a huge dataset of handwriting examples. That is millions of samples of handwritten text + information about what the written text says in every example).

    This is why the best engines only exists as a service in the cloud. The OCR engines you can install lovely that are acceptable, but far from perfect, are commercial. Parascript FormXtra is one of the better commercial ones.

    The only OCR Engine that's free and really good is Tesseract OCR but it doesn't handle handwritten text.







  • By your reasoning, every single platform should be in the same shitty state of yt

    What comparable platforms are you talking about that is not running ads or have some sort of pay-to-watch?

    If we talk about Twitch and their revenue I can promise you that they would not be very profitable without female streamers dressed sexy that doesn't always play video games.

    We now live in a world where users got used to never have to pay for content or experience. Even though Google makes insane money in different areas the cost for running and developing YouTube is huge. I'm not a fan of ads (I don't see ads when at home because of how I have set up my network) and the subscription plans always seems too pricey for the value I get when using different streaming services

    But all of this doesn't change the fact that even though I don't like ads or paying for content I still haven't come up with a better solution myself.



  • If one video stream to one user uses 128 kilobyte per second out of your 100 megabit internet connection 781 users can watch that stream at the same time. However, the ISP will charge you per transferred gigabyte each month. So let's say that you serve 781 users that video 24/7 in a full month of 31 days ... It will be 100 megabit divided by 8 to get 12.5 megabyte. So it's 12.5 megabyte per second. That's 750 megabyte per minute. That's 45 gigabyte per hour. That's 1 terabyte or day. So around 31 terabyte traffic per month. (If you use this much bandwidth you will get a discount but it's still not going to be

    Now, that's just for 781 simultaneously users.

    What is we need to serve 781000 simultaneous users?

    Now, this far we've only been talking about one video on repeat 14/7. What about 100000 videos and enough programmers and computers to design as system that lets each and every user choose any video whenever they need to? Now you suddenly have thousands of servers and harddisks running in a couple of hundred places on earth 24/7.

    Now this is for you to provide your users 100000 different videos even before you start to pay content creators for their hard work.

    Also, you need to be available 24/7 so now you have to make backups, redundant servers on different location that can take over in case of an accident, dedicated internet connection (being alone on the internet cable is not the same as sharing it with 100 other sites) and a whole lot of other things you need to take care of.

    What about offering the 500 million videos YouTube offers their users?

    ... and all of this cost is paid out of your pocket?