Really appreciate you putting chat messages in subtitles

thank you for using ethically sourced lena, it wouldn't have been nearly as funny

Watson-Watts said it like this: "Give them your third best solution. The second best comes to late, the best never comes". He was a pioner in radar development before WWII.

All of the failures before you finally succeeded were hilarious. One suggestion there is to make a video of all the intermediate images. I'll have to remember that about the chat, because sometimes they say useful things. As far as memmove() and memcpy(), I think the biggest reason that so many people have problems with them is because C tries to make arrays friendlier to use by scaling your indices by the sizeof the base type of the array, yet when you call one of these functions they expect the element count in bytes. It might be more useful for people if we had an arrmove() or arrcpy() function that took the base pointers to move/copy to/from and then expected an element count. I suppose you could just wrap the calls to memmove() and memcpy() with a macro and call them with sizeof(*a)*(elem_count) for the last argument.

You really need to have balls to refactory code like you does lol I really like to commit a save point first, because if I get lost during the refactory I don't like to depend on undo. But I think this is a skill issue lol thanks for the great content Mr. Zozin!

That's really awesome how well that works for something so simple, thanks for the video!

Yo... That algorithm is almost a thunderstorm simulator with all of that energy and path of least resistance talk.

@TsodingDaily if you liked working with energy on an article, I have a fond memories of implementing "As-Rigid-As-Possible Surface Modeling" by Olga Sorkine and Marc Alexa. It’s easy to find online. It uses another energy minimization approach but to compute animations. It works in 2D and 3D and, from what I remember, is not _that_ hard to implement (you might need (or implement yourself) a matrix algebra library though). Thanks again for you videos, always enjoy them !

this was a great video. really interesting concept for a programming session.

Hi, always a great content... Have you ever done or thought about doing anything related to LPR? License Plate Recognition?

Tsoding myaahn, u r an inspiration. *salute*

Every time I think dead internet theory isn't real I get reminded of your twitch chat

I’ve used this feature in photoshop (and kept the result) like… maybe once in 10 years lol. It is a cool algorithm though.

Damn ethically sourced Lena looking even better! Looking forward to the video game in ada video!

The twitch subtitles is a really nice touch

I beg to differ this algorithm "works perfectly on faces" this is amazing!!🤣

1:23:22 that formula is wrong. If the value for dx=-1 is smaller you never actually check for dx=+1. Also, I think the direction of the seem could be already computed in the dynamical programming step for each pixel. Another cool idea is to do the re computations only locally along the seem, since most the picture haven't changed.

QOI (Quite Okay Image) and QOA (Quite Okay Audio) formats implementations when?

Why don't you code GIT (the notorious version control system) from scratch in C? It'd be a very interesting topic.

I read the description as "Sean Carving".and thought "R.I.P. just killed with code".

I love this

I wonder if it could be extended to add space, and if it could also be extended to add/reduce space vertically.

How do you find so many cool papers😮

This is super cool

that was pretty classno bratan. ya learning ochin good s taboy.

1:51 Man being able to choose a topic for recreational coding and having to worry about it using AI is truly a sign of the times

@47:30 shouldn't that be "mn + (mat[i] / (mx - mn))" ? you're including the min offset into the division when it shouldn't

Damn, my current hw for computer vision is exactly this topic.

Do you have a video with Mutex Locking | Thread Synchronization? I need that for my project

01:07:20 out of bounds 0.0 is not correct when: grad is 1x2 table: { {1.0}, {1.0} } after your algorithm dp is: { {1.0}, {1.0} } but dp should be: { {1.0}, {2.0} } edit: chat noticed 02:24:46

when are you doing it in assembly sir ?

20:41 oh I remember the rant from Acerola, I now always use Oklab instead of that shite

Excellent work. I've not watched all of it yet but will have to to get my head round your awesome code. Dropping the scale down factor from 2/3 to 1/2 then using it both horizontally and vertically has most hilarious looking effects on pompous right wing politicians.

@TsodingDaily 42:23 I'd recommend to just assign the first element's value to min and max

May I ask why you prefer C over Rust?

Thank you....

The problem is that the way you calculate the gradient. The selection then really loves the left most column.

So why FLT_MAX instead of 0.0 was a problem? 0.0 would make a low gradient on the borders making the algorithm prefer borders ?

Well no technically 2005 was not before AI, AI technology dates back to the '70s and '80s

ah so this is how those big eyes memes are made

also now make reverse so it adds pixels and you have stable diffusion

jpeg encoding would be enough but way more faster. png is for pictures with large regions of same pixels like screenshots

How can i join the discord server?

I'm going to rewrite in rust.

15:44 I still believe in an pure algorithmic approach than statistical/AI. AI is just brute forcing things, it not "intelligent", its actually very stupid. I believe in that so much that I actually have a computing vision company where I'm betting on this. The YOLO algorithm on OpenCV still bets on that, it is not AI, one the best algorithms out there outperform the AI ones, and it is used in the industry because its reliable, AI is not reliable, just like humans. We want computers to be machines, not humans. (although I have an algorithm that I hand optimized that outperforms the last YOLO, its obviously proprietary because I need to pay my bills, sorry)

The reason russian government likes Tsoding, is because of him shit talking the NSA.

23:25 0xFF000000|(value*0x010101)

if (working) twerking = true;

Привет человек я ваш поклоник 😊. Язык 👅 что ты придумал думую лучше не использовать его без сокращений, в цели соответствии с русским. А так ты крут, подержать тебя у меня пока не получится держись.

best wasted 2hours 36minutes, was all worth it for the last 2 minutes, 2:34:36

zozin nooo don't die тwт

I hate to double-comment, but this in action makes me remember when Twitter tried to add something like that with fancy AI stuff - if I recall correctly, someone tested with images containing Obama and Biden and the AI was always removing Obama from the picture. This algorithm is definitely much cheaper than any "modern AI" and works much better.

I like my data dream linux command (dd if=/dev/null of=/dev/sda)

Baba C# ......

First

4 views in 37 seconds. Damn you fell off

52:10 I also say, it is not about programming language, it is about well written libraries and basic knowledge of the problematics

CASoding