Download "Я сделал ИИ для Aim Lab и она его уничтожила :3"

Please wait. We're preparing links for easy ad-free video watching and downloading.

Previous video: НА МУВИК | КС ГО Next video: S1mple mode | CS:GO Edit

0:00

Идея

0:25

Что такое Aim Lab и зачем бот?

0:55

Начинаем писать код (захват экрана)

1:35

Первая задача (определение целей)

1:55

Варианты решения задачи

2:29

Цветовые маски

2:55

Совет про скейлинг HSV

3:30

Проверяем код (маску)

4:03

NMS и объединение по пересечениям

4:32

Мой рекорд в аим лабе

4:50

Вторая задача

5:20

Как работает мышь на уровне ОС и в играх

6:42

Как играет смещает прицел

7:10

Проблема перемещения мыши

7:48

Разбор проблемы

8:37

Следим за руками (считаем углы)

10:25

Про машинное обучение

12:27

Конвертируем координаты из 2D в 3D (пытаемся)

13:35

ПАМАГИТЕ

13:57

Костыли наше всё!

14:49

Первый боевой тест

15:28

Второй боевой тест

15:57

Итоги

S1mple mode | CS:GO Edit

S1mple mode | CS:GO Edit

Channel: Ins1derFT

НА МУВИК | КС ГО

НА МУВИК | КС ГО

Channel: Darlin

Жесть вспомнил про CS 2

Жесть вспомнил про CS 2

Channel: Drag Gray

M0NESY ЗАШЁЛ В НАПАРНИКИ С ДРУГОМ / МОНЕСИ ИЗДЕВАЕТСЯ НАД ПРОТИВНИКАМИ И БЕРЕТ ИХ НА НОЖ (CS:GO)

M0NESY ЗАШЁЛ В НАПАРНИКИ С ДРУГОМ / МОНЕСИ ИЗДЕВАЕТСЯ НАД ПРОТИВНИКАМИ И БЕРЕТ ИХ НА НОЖ (CS:GO)

Channel: FASTOV CS 2

СДЕЛАЛ НОВЫЙ АНТИ-АИМ ДЛЯ CS:GO (репорт обеспечен)

СДЕЛАЛ НОВЫЙ АНТИ-АИМ ДЛЯ CS:GO (репорт обеспечен)

MONESY УДИВЛЯЕТ ДЕВУШКУ В ПАТИ СВОИ ЖЕСТКИМ ЭЙСОМ НА FACEIT!

MONESY УДИВЛЯЕТ ДЕВУШКУ В ПАТИ СВОИ ЖЕСТКИМ ЭЙСОМ НА FACEIT!

Channel: CS GO TWITCH

aim lab

аим лаб

кс

source

сурс

чувствительность

sensitivity

cs go

кс го

king dm

fispect

кинг дм

нейросеть

ии

я сделал ии

я сделал игру

opencv

компьютерное зрение

computer vision

боты

браузер

counter strike

python

нейроэволюция

фид форвард

сделал игру

unity

web gl

кс:го

ксго

s1imple

симпл

navi

хауди

хо

хауди хо

00:00:02

told myself that if I don’t make a Bot

00:00:05

that will be at least in the top 10 And they have

00:00:08

a lab, then I won’t upload the video to the

00:00:10

channel, and since you can watch it now, I

00:00:13

must assume I have something

00:00:15

all- it turned out like this Today I

00:00:18

will tell you how I made one of the

00:00:20

most complex and coolest bots for aim

00:00:23

in just a few days. Well, if

00:00:25

someone suddenly doesn’t know what A im paw is,

00:00:28

then this is such a very popular game,

00:00:30

especially among e-sportsmen who

00:00:33

can train and improve your

00:00:35

aiming skill or aiming in the game

00:00:38

Well, the reason why I decided to make a

00:00:40

Bot is essentially only one. It’s stupidly fun,

00:00:43

interesting, and one more thing. To be honest, I

00:00:46

wanted to get into the top 10, and since there are still

00:00:49

four situations there,

00:00:51

it’s worse for me I definitely won’t do

00:00:55

this. To begin with, as usual, I wrote the

00:00:59

simplest code for capturing screenshots from the

00:01:01

screen. Because my bot will

00:01:03

use one of the areas of

00:01:05

artificial intelligence for its

00:01:07

work, namely computer vision, it’s

00:01:10

kind of like human vision, only it’s

00:01:13

computer vision, that is, in fact, the bot will be

00:01:16

see the same thing as a person and

00:01:18

somehow the readers will write the

00:01:20

game process into memory, we won’t have it at all; we

00:01:23

’ll only have a picture from the game on our hands. Well,

00:01:26

as you can see, capturing from the start produces

00:01:28

approximately 50 FPS at Full HD resolution and

00:01:32

this is very cool However, we are

00:01:35

not processing this image in any way yet and

00:01:37

perhaps the first task that we

00:01:40

now face is to somehow determine the

00:01:42

position of the goals in the picture, while as

00:01:45

you see our goals - these are blue balls,

00:01:47

their color does not change in any way during the game,

00:01:50

which of course makes the task much easier for us,

00:01:52

so

00:01:54

why not do it personally?

00:01:57

Three different options immediately come to mind: we

00:02:00

could use the

00:02:01

After hotkey program and its imatech method,

00:02:04

which allows you to find a given picture on the screen; you

00:02:05

can also create your

00:02:08

own hard ask for the

00:02:10

Open CV computer vision library and also,

00:02:13

in principle, no one is stopping you from

00:02:15

training an entire neural network to

00:02:17

detect objects, in this case

00:02:19

blue balls, however, I successfully sent all three options

00:02:22

to the landfill Since I had already

00:02:24

decided in advance to use a more

00:02:26

adequate approach, namely color

00:02:29

masks. Well, the essence of this method is

00:02:31

that what we First,

00:02:33

we convert the image into the

00:02:35

hsl color space and then determine the

00:02:38

lower and upper boundaries of the color mask

00:02:40

in order to ultimately exclude all colors from the image

00:02:43

except those that fall into our

00:02:46

color mask and And you don’t have to

00:02:48

worry. If you don’t understand anything, I didn’t

00:02:51

understand anything either However, this garbage

00:02:53

works and the only thing that is enough for us is I

00:02:57

want to say that you, like me, need to

00:02:59

be more careful and take into account that the

00:03:01

skelling of the hss in different programs can be

00:03:03

different, for example in Photoshop or Color

00:03:06

pickery, the conditional blue color will be equal to

00:03:08

180 While in the Open CV library

00:03:12

that The Color will be in the value

00:03:14

90 That is why, to determine these

00:03:17

very color boundaries, I used a

00:03:20

small cheat sheet with the designation of

00:03:22

colors in the xss space, respectively, the

00:03:24

blue color of the balls we need is

00:03:27

approximately 90 So if we

00:03:30

run the code now, we will get the following result

00:03:33

As you can see from the picture were

00:03:35

Absolutely all colors except those defined in

00:03:38

our mask are excluded and thanks to this we now

00:03:41

see only balls, which means we can already

00:03:44

draw boxes for detected targets in the

00:03:46

game. It now looks like this,

00:03:48

but this method has one

00:03:52

big drawback, and this is

00:03:54

the amount of garbage or incorrect

00:03:56

definitions for example, here for some reason a

00:03:59

small square was detected inside

00:04:01

another one. Although this should not be the case. In order

00:04:03

to fix this, we

00:04:06

need to somehow filter the detected

00:04:08

boxes. And for this, I personally used

00:04:11

two simple algorithms, the first is nms

00:04:14

or but the maximum is interrogated. And the second is a

00:04:17

union by intersection together This

00:04:20

gave results and now there are

00:04:23

practically no incorrect definitions, which

00:04:25

means that we now have in our hands the

00:04:27

coordinates of the targets in the picture, this is already

00:04:30

something. By the way, when I myself try to

00:04:32

score as many points as possible in Wimlab,

00:04:35

I get a maximum of 60 or 70,000

00:04:39

if any of you played on this map,

00:04:41

then write in the comments How many points

00:04:44

did you manage to score? I think everyone in

00:04:46

the comments will be interested to know,

00:04:49

well, we continue And the next Step in our

00:04:52

reliable, like a Swiss watch in this

00:04:55

regard, is to move the sight in the game

00:04:57

so that it automatically I was aiming at

00:04:59

the balls and I just said

00:05:02

the word sight and not the cursor for a reason. Although it

00:05:06

would seem that there is no difference at all, but

00:05:08

it turns out there is a difference and it is very

00:05:11

big, and this difference is

00:05:13

so big that moving the sight

00:05:16

in the game to the desired target just won’t

00:05:18

work. will always miss

00:05:20

and I will now explain why, look at the

00:05:24

operating system level, the mouse

00:05:26

works in two modes of movement,

00:05:29

namely absolute and relative;

00:05:31

besides, for the operating system there is no such thing

00:05:34

as a pixel; instead,

00:05:36

conventional units called T-shirts and

00:05:40

T-shirts are used, this is a unit that determines the

00:05:42

minimum detectable shift mice, as a

00:05:45

rule, the T-shirt is equal to 1/10 mm, which in

00:05:48

practice moves the mouse cursor by 2

00:05:50

pixels for each unit of this same

00:05:52

T-shirt. Next, you need to understand such a thing that

00:05:56

in games, in order for the sight

00:05:58

to move freely 360 degrees And

00:06:01

this would not move the mouse cursor beyond borders of the

00:06:04

game window, as a rule, two

00:06:07

approaches are used, the first approach involves

00:06:09

constant teleportation of the mouse cursor from

00:06:12

one edge of the screen to the other. When the mouse

00:06:15

approaches it, the second one works even

00:06:17

more rigidly and simply blocks the

00:06:20

mouse cursor at one point. And between frames it

00:06:23

simply detects the cursor displacement for

00:06:25

detection. Well, It is also necessary

00:06:28

to understand that the world in the game is a 3D

00:06:31

space, if on your desktop you

00:06:33

only have two axes, these are X and Y. Who in the game

00:06:36

already has three of them because the game is like a three-dimensional one,

00:06:39

it goes without saying x y z and this is how to

00:06:43

shift the aim in the game game

00:06:46

the engine converts mouse movement into so-

00:06:48

called angles, and not to mention the fact that the

00:06:51

player’s field of view itself is in three-dimensional

00:06:53

space, but how is it limited by these

00:06:56

same angles, which are also called

00:06:58

fullv or fieldofue or viewing angle

00:07:01

Well, why am I actually doing this? I’m

00:07:04

explaining everything now. the point is that if we

00:07:08

found a ball in the picture at

00:07:10

certain coordinates, then we can’t

00:07:12

just move the cursor along these

00:07:14

coordinates and expect the sight in the game

00:07:17

to target this ball.

00:07:19

And it doesn’t work like that, and that’s all because

00:07:22

Aim paws like almost everything other

00:07:25

games Shooters use the second approach to

00:07:28

detect mouse movement, that is, the game

00:07:30

simply blocks the cursor at one point and

00:07:33

it does not move from there. Or, to

00:07:36

put it in simpler words, they don’t even

00:07:39

care what you are there or where you

00:07:40

move it, it completely ignores the

00:07:43

absolute movement of the mouse at all

00:07:46

[ __ ] it

00:07:48

and probably the only game I

00:07:51

know in which the absolute placement method works is

00:07:55

counter-strike Global Offensive Although I

00:07:57

must admit I’m very surprised Why

00:07:59

Valve still hasn’t closed this shop

00:08:02

here the Source game engine reads the

00:08:05

absolute movement of the mouse and

00:08:07

converts it into angles for you turning the

00:08:09

sight in the game However, we are not in

00:08:12

cs:go now and therefore such blessings of humanity

00:08:14

as absolute movement of the mouse are

00:08:17

not available to us. This means that we can

00:08:19

only move the mouse relatively and

00:08:22

at the same time we still need to figure out how to

00:08:24

convert the mouse coordinates from 2D

00:08:27

space into the angles of rotation of the sight in

00:08:29

3D space and it seems like the task

00:08:32

sounds easy But in reality it’s not so

00:08:36

simple but Let’s start with the

00:08:38

simplest thing, let’s determine the angles and the actual

00:08:41

value of the pixels based on 360

00:08:43

degrees. Later for this I went into the

00:08:46

game settings and looked at what is currently

00:08:48

installed, then I went and registered

00:08:51

We will need it in the code for

00:08:53

calculations.

00:08:54

We also need to go into the game and define a

00:08:57

certain focus point here, and at

00:09:00

this moment carefully watch your

00:09:02

hands. We know that the game resolution is

00:09:05

now 1920 by 1080 pixels, with

00:09:08

horizontal off V equaling 106 26 A

00:09:11

vertical 7374 This means the rotation angle

00:09:15

from the focus point to the screen border

00:09:17

is 106.26 / 2 or

00:09:21

53.13°. Now through the code we move the mouse with

00:09:24

different values until the

00:09:26

crosshair hits the screen border from the focus point

00:09:28

and this value I

00:09:31

got equal to 2.420 pixels, in other

00:09:34

words we need to move the cursor in

00:09:36

relative mode by 2420 pixels

00:09:39

to get a shift of the game sights by 53

00:09:42

and 13 °. This means we can now simply

00:09:45

divide

00:09:47

2.420/53 13 and get a value of

00:09:50

45.54 pixels which will be the

00:09:53

value of pixels per 1 degree if we

00:09:56

multiply it by 360 we get

00:10:00

16400 or the required movement in

00:10:03

pixels to shift the sight a full

00:10:05

360 degrees and we can it’s even very

00:10:09

easy to check this if we write a

00:10:11

shift of 180 degrees in the code and in the game. As

00:10:14

you can see, the sight will also shift

00:10:16

each time exactly 180 degrees

00:10:19

horizontally, which means that our

00:10:22

calculations turned out to be correct and we solved the

00:10:24

first half of the problem. In general, this is a

00:10:27

very cool solution You encounter such problems in code

00:10:29

especially often

00:10:31

in machine learning. And by the way, if

00:10:35

you like the field of machine learning and

00:10:37

want to develop in this direction,

00:10:39

then pay attention to the profession of a data

00:10:42

scientist; in short, this is someone who builds

00:10:44

mathematical models and algorithms that

00:10:46

simplify solutions to problems for business and

00:10:49

users, at the same time, learning machines are

00:10:51

actively used and in games, take for

00:10:54

example even our current implementation

00:10:56

of aimbota for the imlab game, here we used

00:10:59

computer vision, which is one of the

00:11:02

many branches of machine learning,

00:11:04

so to become a specialist you

00:11:07

need not just to study, but also

00:11:09

to practice a lot and in The

00:11:11

online service of digital professions Yandex

00:11:13

workshop will help you with this and they give you the opportunity to take the

00:11:16

water part of the course for free, while

00:11:19

Yandex workshop has three courses

00:11:21

Data Sansa and all of them are for beginners

00:11:24

Data Science specialist - this is a basic

00:11:26

course where you have everything to master a new

00:11:29

profession here you will Learn the basic

00:11:30

concepts of data analysis Detoscience specialist

00:11:34

plus for those who want to

00:11:36

dive deeper into the machine and training and

00:11:39

data analysis and data science Booth camp - This is a

00:11:42

program for those who want to

00:11:44

quickly master the profession And depending

00:11:46

on the chosen program you will complete From

00:11:49

16 to 25 educational projects and two real

00:11:53

projects from the order of subscribers and in the end

00:11:55

you will have a portfolio for

00:11:57

employment; training allows you to

00:12:00

become a Junior Data Science specialist from scratch;

00:12:02

master pyton data analysis and

00:12:05

machine learning; at the same time, already at the start of your

00:12:07

career, salaries start from 60 thousand and

00:12:10

increase with experience, follow the

00:12:13

link under the video or scan the QR code from the

00:12:15

screen and start learning for free. And

00:12:18

using the promotional code hauti, Yandex workshop has

00:12:20

prepared for you a five

00:12:22

percent discount on the full course until December 24.

00:12:26

Okay, then we need to somehow,

00:12:30

as I already said, convert those same

00:12:33

coordinates of detected targets from 2D

00:12:35

space to rotation angles in 3D

00:12:38

space, and to be honest, I have no

00:12:41

idea how to do this,

00:12:51

this is transferring values through percentages. That

00:12:55

is, for example, if the conditional shift of

00:12:57

detection by X is 10 out of 100,

00:12:59

then it is 10 percent and since we

00:13:02

it is known that half the screen is 53 and 13

00:13:05

degrees. Then it’s enough just to get

00:13:07

10 percent of this value and we

00:13:10

’ll get the value of the same shift

00:13:12

but in degrees. Unfortunately, this

00:13:16

sounds beautiful only in theory; in practice,

00:13:18

the sight often goes past the target, although not

00:13:21

very much the second thing I came up with was to

00:13:23

actually make a table for

00:13:25

converting one coordinate grid

00:13:27

to another. But firstly, I was too lazy to

00:13:30

program it. And secondly, it’s still a

00:13:32

crutch, although it’s quite interesting, and I’ll

00:13:34

honestly tell you, hand on heart,

00:13:38

I tried to google it, and even more than

00:13:41

one day, but I couldn’t find at

00:13:43

least some useful

00:13:45

information on how to do this. Therefore,

00:13:48

if any of you are rummaging around or have

00:13:51

ideas, welcome to the comments

00:13:54

enlarged, I don’t know how to solve this yet

00:13:57

problem Well, since I need at least a

00:13:59

slightly working solution, I settled on the

00:14:02

first option, adding a crutch to it in the

00:14:05

form of a correction table so that at

00:14:07

different angles of rotation the sight would not recoil

00:14:10

so much And although this, frankly speaking,

00:14:12

looks funny in the game, the aiming

00:14:15

finally worked and the sight it is almost

00:14:18

always correctly aimed at the target and for

00:14:21

convenience, I also added in the code

00:14:23

the display of red dots, that is, the places

00:14:25

where the bot is trying to aim, and ran

00:14:28

the very first test. As you can see,

00:14:30

aiming, in principle, works well

00:14:32

and the bot shows good results.

00:14:34

However, it sometimes misses and this

00:14:38

is read out in the game as a decrease in points,

00:14:40

so I quickly added in the process of

00:14:42

detecting color under the sight and if

00:14:45

the sight was suddenly updated incorrectly, then

00:14:48

shooting does not happen Well, then I

00:14:51

finally launched the very first combat of

00:14:54

our Bot and As you can

00:14:56

see for yourself, it copes very

00:14:59

well on the first try in 60 seconds the bot

00:15:02

managed to score as many as 138 thousand points and

00:15:06

immediately take 22nd place in the ranking, but as

00:15:09

you remember, my goal is the top 10 and

00:15:12

if I don’t make a Bot that can

00:15:14

take the top 10, then I won’t release the video,

00:15:17

so I’ll continue and made

00:15:19

several improvements to the code, the most important

00:15:22

of which is the detection of the nearest

00:15:24

target, so it will be easier and faster

00:15:26

to aim, and finally, the final

00:15:30

combat test of the Bot looks like this

00:15:33

from here By the way, we can also

00:15:35

conclude that a person is simply not capable of

00:15:38

scoring more than 100,000 points, and ginger at

00:15:40

least on this map And this

00:15:43

means only one thing: the leaderboard in the top is

00:15:45

full of bots and cheaters, and my bot,

00:15:49

although still very crude, was able to score

00:15:51

143,000 points and take the Top 10 on

00:15:55

this Wimlap map. It should be

00:15:58

understood that this is not a bot in in the classical

00:16:00

sense, since our code does not read or

00:16:03

write the processor in the game memory, and all

00:16:06

that it has on hand is just a

00:16:08

picture from the game, and all that we use

00:16:11

to operate the Bot is computer vision,

00:16:14

which is one of the bus

00:16:16

learning farms and is also widely used in

00:16:19

artificial intelligence about and I haven’t

00:16:22

forgotten yet if anything I’ll leave the source code for this

00:16:24

lesson in the description under the video,

00:16:27

maybe someone will be interested in how you

00:16:29

poke it or maybe even improve it,

00:16:32

otherwise I’m waiting for your royal likes

00:16:35

of course subscribe to the channel and

00:16:37

turn on the notification bell and

00:16:39

also write comments, at least

00:16:42

because I read everything. And it also helps a lot to

00:16:44

promote the video on YouTube. In short, you

00:16:47

understand good luck and always remember the more

00:16:49

difficult the task, the more experience it

00:16:52

will bring you

00:16:54

[music]

Description:

Делаем ИИ бота для игры Aim Lab. Компьютерное зрение в деле :3 https://practicum.yandex.ru/data-scientist/ — выберете для себя подходящую программу курса «Специалист по Data Science» в Яндекс Практикум. Промокод "ХАУДИ" на скидку 5% до 24 декабря 2022 года. 🆇 Исходный код бота 🆇 https://github.com/Priler/aimlabbot 🆇 Главы 🆇 0:00 - Идея 0:25 - Что такое Aim Lab и зачем бот? 0:55 - Начинаем писать код (захват экрана) 1:35 - Первая задача (определение целей) 1:55 - Варианты решения задачи 2:29 - Цветовые маски 2:55 - Совет про скейлинг HSV 3:30 - Проверяем код (маску) 4:03 - NMS и объединение по пересечениям 4:32 - Мой рекорд в аим лабе 4:50 - Вторая задача 5:20 - Как работает мышь на уровне ОС и в играх 6:42 - Как играет смещает прицел 7:10 - Проблема перемещения мыши 7:48 - Разбор проблемы 8:37 - Следим за руками (считаем углы) 10:25 - Про машинное обучение 12:27 - Конвертируем координаты из 2D в 3D (пытаемся) 13:35 - ПАМАГИТЕ 13:57 - Костыли наше всё! 14:49 - Первый боевой тест 15:28 - Второй боевой тест 15:57 - Итоги 🔵 Наш TELEGRAM: https://t.me/howdyho_official Наш ВК: https://www.vk.com/howdyho_net Сотрудничество https://vk.com/topic-84392011_33285530 💗 Музыка предоставлена YouTube Audio Library.

Preparing download options

Popular

HD video

Only sound

All

* — If the video is playing in a new tab, go to it, then right-click on the video and select "Save video as..."

** — Link intended for online playback in specialized players

Questions about downloading video

How can I download "Я сделал ИИ для Aim Lab и она его уничтожила :3" video?

http://unidownloader.com/ website is the best way to download a video or a separate audio track if you want to do without installing programs and extensions.
The UDL Helper extension is a convenient button that is seamlessly integrated into YouTube, Instagram and OK.ru sites for fast content download.
UDL Client program (for Windows) is the most powerful solution that supports more than 900 websites, social networks and video hosting sites, as well as any video quality that is available in the source.
UDL Lite is a really convenient way to access a website from your mobile device. With its help, you can easily download videos directly to your smartphone.

Which format of "Я сделал ИИ для Aim Lab и она его уничтожила :3" video should I choose?

The best quality formats are FullHD (1080p), 2K (1440p), 4K (2160p) and 8K (4320p). The higher the resolution of your screen, the higher the video quality should be. However, there are other factors to consider: download speed, amount of free space, and device performance during playback.

Why does my computer freeze when loading a "Я сделал ИИ для Aim Lab и она его уничтожила :3" video?

The browser/computer should not freeze completely! If this happens, please report it with a link to the video. Sometimes videos cannot be downloaded directly in a suitable format, so we have added the ability to convert the file to the desired format. In some cases, this process may actively use computer resources.

How can I download "Я сделал ИИ для Aim Lab и она его уничтожила :3" video to my phone?

You can download a video to your smartphone using the website or the PWA application UDL Lite. It is also possible to send a download link via QR code using the UDL Helper extension.

How can I download an audio track (music) to MP3 "Я сделал ИИ для Aim Lab и она его уничтожила :3"?

The most convenient way is to use the UDL Client program, which supports converting video to MP3 format. In some cases, MP3 can also be downloaded through the UDL Helper extension.

How can I save a frame from a video "Я сделал ИИ для Aim Lab и она его уничтожила :3"?

This feature is available in the UDL Helper extension. Make sure that "Show the video snapshot button" is checked in the settings. A camera icon should appear in the lower right corner of the player to the left of the "Settings" icon. When you click on it, the current frame from the video will be saved to your computer in JPEG format.

What's the price of all this stuff?

It costs nothing. Our services are absolutely free for all users. There are no PRO subscriptions, no restrictions on the number or maximum length of downloaded videos.