Hey, thank you so much for your comment! It really made my day!

@@moisesdepaulodias7980 you can help me pls

Agreed. This is legendary

Hi there, thank you for your comment! A few months ago I was working on a bot for Ragnarok Online, but I was having a hard time finding online content showing how to create a object detector and make the detection results available in Python. So, when I created my one, I decided to try making a video about it (this video), I didn't expect to have such a good reception. I'm very happy that it worked and was useful for you!

@@moisesdepaulodias7980 that was what I wanted to do. I started with template matching from OpenCV, but it didn't worked quite good and I started looking through using OpenCV for object detection but couldn't find something so easy as what it is in this video. The only issue that I encountered, is that the model is a bit delussional and makes detection of other objects, but I will train another model after I finish with implementing all the things for my bot. Thank you very much. Also, do you know any resources for learning OpenCV. I'm a programmer, but I haven't looked through a lot of resources, what do you recommend for starting up with OpenCV

Thank you so much for the feedback! I'm really glad the tutorial was helpful for you!

Thank you so much for your comment!!!

Thank you so much for your comment! It really means a lot to me!!!

@@moisesdepaulodias7980 You're welcome. I get a warning message in the last cell when I attempt to train my model, it reads: "Warning: in txt-labels class_id=-1 >= classes=1 in cfg-file. In txt-labels class_id should be [from 0 to 0] Should I Ignore this? Eventually when it is done training i'll be able to see if it worked or not anyways.

Hey, thanks for your comment! I'm a little busy with my job right now, but I want to make more videos (mostly about other approaches without using machine learning, that use far less computer resources and in some cases they have the same result)

Hey man, thank you so much for your comment!!! I wasn't expecting such good feedback, but I am truly happy that my video is helping you out!!!

Hey!! I am happy to know that it worked for you!!!

Hey, thank you so much for your comment!!

Hey, thanks for your comment man!!

opa, muito obrigado mesmo pelo comentário mano, de verdade!! Fico muito feliz em saber que vc gostou do vídeo!!

Opa, muito obrigado pelo elogio mano!! De verdade mesmo!! Vou tentar trazer mais tutoriais como esse. Abraço!!

Hi there. I had the opportunity to try to help one person with the same problem in a Discord call, and I believe that this is related to some anti-cheat of your game. Because in this Discord call, the code worked in some games, but in others it generated black screens (I also suspect that games within Steam may generate black screens as well). I suggest you try with other games to see if that is your case. For me, it worked with the games Terraria, Ragnarok Online, Diablo 2, and Fruit Ninja using the BlueStacks emulator. I am trully sorry I can't help any further, but I didn't encounter this problem while creating the tutorial.

You're welcome! :D

Opa, muito obrigado pelo seu comentário man, de verdade!! Como falei em algum comentário anteriormente, eu tentei achar um conteúdo assim como nesse vídeo, mostrando a criação de ponta a ponta de um sistema de detecção de objetos disponibilizando as coordenadas em python, mas não achava nada, daí qnd fiz o meu tentei fazer esse vídeo, mas não esperava uma recpção tão boa! Fico muito feliz que vc tenha gostado man!!!

Muito obrigado pelo comentário Rodrigo!! De verdade!!!

Hi there, thanks for your comment! Only step 3 (Model Training) is performed in Colab, all the other steps are using a local environment. I would not recommend using CPU for training your model for two reasons: first, it will use almost 100% of your CPU while your model is training, and second, it will take far more time than using a GPU. If you are facing GPU limitations, I would suggest that you create a new Google account and access Colab using this new account. This way, you will have a new GPU usage quota. Can you try that and see if it solves your problem?

@@moisesdepaulodias7980 Hello, again :) well I managed to pass the model training step, but I got stumbled across the following error while running the detector --------------------------------------------------------------------------- error Traceback (most recent call last) Cell In[20], line 8 5 weights_file_name = "yolov4-tiny-custom_last.weights" 7 wincap = WindowCapture(window_name) ----> 8 improc = ImageProcessor(wincap.get_window_size(), cfg_file_name, weights_file_name) 10 while(True): 12 ss = wincap.get_screenshot() Cell In[19], line 11, in ImageProcessor.__init__(self, img_size, cfg_file, weights_file) 9 def __init__(self, img_size, cfg_file, weights_file): 10 np.random.seed(42) ---> 11 self.net = cv.dnn.readNetFromDarknet(cfg_file, weights_file) 12 self.net.setPreferableBackend(cv.dnn.DNN_BACKEND_OPENCV) 13 self.ln = self.net.getLayerNames() error: OpenCV(4.8.0) D:\a\opencv-python\opencv-python\opencv\modules\dnn\src\darknet\darknet_importer.cpp:217: error: (-212:Parsing error) Failed to parse NetParameter file: yolov4-tiny-custom_last.weights in function 'cv::dnn::dnn4_v20230620::readNetFromDarknet' Thanks for caring

Hi there! Thanks for your comment! I am using the integrated GPU of my laptop, it's a Radeon 520 I believe.

Hi there! Thank you for the suggestion! I will consider it for my next videos.

Hi there! Thanks for your comment!! Sorry but I didnt understand your question, you need to detect only part of the browser window?

@@moisesdepaulodias7980 Hi! thanks for your reply:D. I already solved it. I had my problems finding the right window_title I think, but then it did work somehow! Finished the whole project, it was amazing, functions perfectly!

@@Noelnnn Great!! I am happy to know that it worked!!!

same issue

same issue, stardew valley, ark, delta force2, spotify... all of them gives black screen

Same 😢

Hi there. I had the opportunity to try to help one person with the same problem in a Discord call, and I believe that this is related to some anti-cheat of your game. Because in this Discord call, the code worked in some games, but in others it generated black screens (I also suspect that games within Steam may generate black screens as well). I suggest you try with other games to see if that is your case. For me, it worked with the games Terraria, Ragnarok Online, Diablo 2, and Fruit Ninja using the BlueStacks emulator. I am trully sorry I can't help any further, but I didn't encounter this problem while creating the tutorial.

Hello! I dont know how to put the bounding box in the game screen. About the movement automation, that is possible to do! I created a new video where I show how you can automate actions based on the detections, you can see more details on the video: kzfaq.info/get/bejne/ncp5iauVt8jMf6c.html

Hello. Did you make any changes to the code in the fourth notebook other than the window name? And did you copy the weights file correctly after training in the third notebook? Also, is there any error message on the output of the cell "# Run this cell to initiate detections using the trained model." ? Please send me a message on Discord so we can investigate together. My discord is moises_dias.

update: I just realized the same is happening for: 1_generate_dataset - Window Capture

Found the problem, for anyone with the same issue: The problem was running the game through steam

Hi there, thanks for your reply! You can only draw the detections on the OpenCV window. But if it's affecting the FPS you can stop showing the OpenCV window by removing the line "self.draw_identified_objects(img, coordinates)". This window is useful only to check if your model is well trained, as you already will have a python dictionary with the coordinates to automate your character.

Hi there, you can use the 'Jupyter' extension in VSCode to run notebooks. I used the Jupyter notebook web interface for simplicity.

me too! I started using some simple detectors (using colors, and shapes) until I needed some more complex detectors for detecting multiple different monsters at the same time (I used diablo 2 in this video but I was actually using this bot in ragnarok!)

I just started to write a CV farming bot (hopefully > macro bot), googled YOLO and found your video 🎁 I was wondering if I play RO under different resolutions, do I have to retake the screenshots and relabel all the data? Maybe I can just apply some image transformation (scaling, distortion) to save some time? Also, we already have the monster image data in data.grf (like d2data.mpq). Is it a good idea to extract the grf data to be used for training? The actual game will zoom in and out. Also, the grf data does not include map details. So I guess that's a no. Do you know if it's possible to run the computer vision algo while the game is minimized / in the background, and send mouse clicks / keys to that window? It seems CV-based bots is limited by this, where you can only run 1 bot at a time (without a VM)

Hi there! I created a continuation for this tutorial where I use pynput to automate actions in the game Fruit Ninja. With the same approach I used in these videos, I believe you can create a bot for these two games! Here's the link to the second video: kzfaq.info/get/bejne/ncp5iauVt8jMf6c.html If you have any questions you can contact me and I will do my best to assist where I can.

opa tmj irmão! obrigado pelo comentário!!

Пожалуйста!

Hi there, I recently helped another person with the same problem, in his case the problem was the game he was trying to detect objects, but in your case you said you were using bluestacks, so I dont know what is the problem.. can you call me on discord for us to search together? my user is moises-dias

found out: my pc had a 125% zoom

@@yatomi1012 Hi There! Sorry for the long time to answer, but I am happy that you solved the problem!!

Hi there, can you send me a message on linkedin with more information about the error? my linkedin profile is in the description, I will do my best to assist where I can.

He is probably running the game in full screen mode, where the winapi is not able to capture since the data is stored internally. If you switch to windowed mode, it probably will work.

Opa, tudo bom? Cara, esse modelo serviria sim, mas como você só quer detectar um icone (ponto de exclamação) da pra voce usar formas bem mais simples (usar o Template Matching do OpenCV) e que seriam efetivas e usando muito menos processamento. Da uma olhada nessa playlist, sugiro vc tentar essa abordagem dela, acho que vai super resolver para você: kzfaq.info/get/bejne/gcuTgM99utvNdWg.html e se precisar de alguma ajuda só chamar man, valeu!!

Hi there! Right now I am working on a playlist of machine learning on Google Cloud, but in the future I will make a video with a newer version of yolo, thanks for the suggestion!!

Hi there, for tasks such as collecting resources (assuming that the sprite or the image of the resources does not change), normally using template matching with OpenCV is a good enough approach. It's also less CPU-intensive than a real-time deep learning model. You can see more in this playlist: kzfaq.info/get/bejne/gcuTgM99utvNdWg.html

Hi there, well, I think so, depending on the graphics you need far more than 80 images.

@@moisesdepaulodias7980 ok will give it a shot, thanks!

Hi there! Hmm, I haven't encountered this problem, but could you call me on Discord so we can try to fix it? My username is moises_dias. I'll do my best to help you.

@@moisesdepaulodias7980 thank you I will add you on Discord, and contact you.I have the same profile image on Discord

@@moisesdepaulodias7980 I sent youa friend request on Discord, I have the same profile picture like here. thank you in advance

hi there, can you send me a screenshot of the error in discord? my user is moises-dias, I will try to help you there

Hi there! Sorry, I didn't understand your question.

@@moisesdepaulodias7980 i mean, bot auto attack

resumindo: preciso enviar clicks de mouse em coordenadas relativa a janela alvo e que nao "prenda" meu mouse e que rode em background..... meu objetivo é conseguir isso, ate o momento sei como fazer mas nao sei como fazer dar certo kkk

opa, fala man, primeiramente, valeu pelo elogio!! e cara, quanto a enviar inputs a partir das detecções, fiz esse segundo vídeo, uma continuação desse primeiro, da uma olhada lá que tem o que você precisa!! kzfaq.info/get/bejne/ncp5iauVt8jMf6c.html

@@moisesdepaulodias7980 vi esse vídeo e a função que eu quero é exatamente dessa maneira porem que eu tenha o mouse "livre" para poder, além de rodar em threads, rodar em segundo plano sem comprometer o uso do mouse em demais ações, no jogo mir4 é estranho, ele recebe input de click porem não na coordenada que pedi e se o jogo não estiver focado o click fica muito superficial, só alguns botões dao realmente ativados....não sei explicar mt bem kk

Hi There! Unfortunately, I didn't tested with a .pt file.

Well, if there's a way to use a machine learning model using JS/Node, I believe it's possible. But I don't have experience with that language. I suggest you look into how to perform the actions I do using Python (like identifying a window, taking a screenshot of the window, using a YOLO model, etc.) but with Node.

My new pc did the same thing Download Python i used 3.10.5 make sure you click add to path during the installation

Hi there. I had the opportunity to try to help one person with the same problem in a Discord call, and I believe that this is related to some anti-cheat of your game. Because in this Discord call, the code worked in some games, but in others it generated black screens (I also suspect that games within Steam may generate black screens as well). I suggest you try with other games to see if that is your case. For me, it worked with the games Terraria, Ragnarok Online, Diablo 2, and Fruit Ninja using the BlueStacks emulator. I am trully sorry I can't help any further, but I didn't encounter this problem while creating the tutorial.

Thanks for your reply I also tried BlueStacks but faced the same problem in my windows 11

Hello! Some games have software to prevent the use of external programs that interfere with the game (anti-cheats). In the game Ragnarok Online, I couldn't use Pynput, so I had to connect an Arduino to control the mouse and keyboard. I haven't tested it in League of Legends, but in your case, I would suggest you look for another way to capture images that the game doesn't block (using OpenCV instead of win32gui, maybe).

@@moisesdepaulodias7980 can you make an example on a game like Waven on steam ? ta show us the game window gestion 😁 you try all my favorite game continue ! 🤟

@@moisesdepaulodias7980 Thank you for your reply. While doing some research, I found a workaround that involves using the following line for the blackscreen debug "windll.user32.PrintWindow(self.hwnd, cDC.GetSafeHdc(), 3)" instead of "cDC.BitBlt((0, 0), (self.w, self.h), dcObj, (self.cropped_x, self.cropped_y), win32con.SRCCOPY)" in your code.

Use pygetwindow instead of win32gui

fala Ricardo, da sim! eu fiz um outro vídeo falando exatamente sobre isso: automatizar ações a partir das detecções, você pode ver nesse link: kzfaq.info/get/bejne/ncp5iauVt8jMf6c.html

@@moisesdepaulodias7980 Sensacional! obrigado Moises! você faz bots para jogos por encomenda?

Hi there, for my goal (detecting the center of the monsters) using bounding boxes was enough. But for more complex applications (like detecting the head of a player in the game counter strike) I believe that an outline trace would be better

@@moisesdepaulodias7980 makes sense buddy

A good reference would be the old Halo game ODDT. One of the best I’ve ever seen.

Hi there, I only used YOLOv4 because I had experience with it in college 😅

Hi there! I believe that this tutorial can also be used to detect objects in a browser window too!

Hi there. I had the opportunity to try to help one person with the same problem in a Discord call, and I believe that this is related to some anti-cheat of your game. Because in this Discord call, the code worked in some games, but in others it generated black screens (I also suspect that games within Steam may generate black screens as well). I suggest you try with other games to see if that is your case. For me, it worked with the games Terraria, Ragnarok Online, Diablo 2, and Fruit Ninja using the BlueStacks emulator. I am trully sorry I can't help any further, but I didn't encounter this problem while creating the tutorial.

Opa, cara, sendo franco, acho que em sites de programadores freelancers vc pode encontrar gente oferecendo esse serviço, eu até poderia vender um bot pra vc mas eu estou com tempo bem limitado, começei um trabalho novo recentemente que ta tomando 101% do meu tempo :(

@@moisesdepaulodias7980 Eu meio que consegui rodar em servidor que tem Gepard Shield 3.0. Porém, ele começa matando normal, mas depois de alguns teleportes ele clicka no monstro e não acontece nada. Eu acho que é por causa do shield. Se vc souber alguém que faça, algum programa bem simples. Por favor, me avisa.

Hi there! Did you add the "yolov4-tiny" folder (created on the file 2_label_dataset.ipynb) to the root directory of your Google Drive?

@@moisesdepaulodias7980 i put yolo 3 in colab then yolo4 in google drive and says no such file, i tried a lot is the same

opa, no colab aparece algum erro ao executar o comando !./darknet detector train ... ? e no final do segundo notebook (rotulando imagens), vc gerou a pasta yolov4-tiny e colocou na raiz do seu google drive?

Hello! I created another video where I show how to automate actions based on detections, one of the actions can be clicking on the screen to walk toward a detected item, you can see that in more detail in the video: kzfaq.info/get/bejne/ncp5iauVt8jMf6c.html

Opa, fala man, cara, acabei fazendo em ingles por que to tentando treinar meu ingles hahahaha, mas quando eu tiver um tempinho livre eu vou adicionar legendas em pt-br então, beleza? Abraço!!!

@@moisesdepaulodias7980 vlw mano e seu ingles ta top man tmj

Hi there! Yes, you can create macros using python after detecting an object in a specific location. You can check more details in this video: kzfaq.info/get/bejne/ncp5iauVt8jMf6c.html

@@moisesdepaulodias7980 can you make for me ?

Same let me know if you find a fix for this

@villocity5794 the game doesn't allow window capturing i had to stream it from steam to a different pc. Or you can use a video capture. Or try OBS

Hi there! Short answer: you need to show all the background so your model will know what he should NOT detect. Long answer: Let's say you have two environments: a forest and a cave, and you want to detect birds in both environments. In the forest, aside from the birds, there are trees and flowers, while in the cave, apart from the birds, there are stones and torches. If you use only pictures of one environment (e.g., the forest) by not labeling the trees and flowers, you are implicitly telling that "the trees and flowers are not birds." However, since you didn't provide your model with pictures of the cave, how can the model know if a stone or a torch is not a bird? The model has never saw these elements and doesn't know if they are or are not birds. This is why it's necessary to show not only what you want to detect but also all the possible backgrounds, so your model knows what it should detect and what it should NOT detect.

Hi there! Can you call me on discord? my user is moises_dias, I can try to help you there

hahahaha valeu pelo comentário man, faço em inglês pra melhorar o meu ingles mesmo

Hello! I belive the problem is that the cell with the WindowCapture was not run. The cell start with "class WindowCapture:". can you try running it and see if it solves the problem?

@@moisesdepaulodias7980 thank you!!

Same issue

Oh boy, if only I had one! I would run three of these detectors at the same time!

opa sim haha

Hi there, can you tell me more details of the error you got?

It is hahaha but it's worth it!

Hello! If you've executed "pip install -r requirements.txt" without any errors, I imagine the issue might be that the Jupyter executable's path is not included in the system's PATH variable. You can copy this exact error text and search it on Google, and you'll find how to resolve this problem. If the issue persists, please let me know in the comments or send me a message.

@@moisesdepaulodias7980 i found the issue ! thanks for help i was have like two python the one i installed with web site and the one with the widows store soo i deleted both and just take the one on web site i was forgeting to make the path variable directly at the installation of python (case that you need to cross)

Hello! In this video, I was clicking on the game window that was on another screen, but I only recorded the window generated by OpenCV. I'm sorry if I gave you that impression, but I didn't perform the action of clicking on the OpenCV window and passing the commands to the game window.

@@moisesdepaulodias7980 So, how do I implement this, can you give me some instructions? Thanks.

Hi there. Probably the folder doesn't have anything else because because the training command didn't work. Please, check if you changed your runtime to use a GPU (I did that at 15:28). Also, compare the outputs of all the commands to my execution in the video. Do you notice any differences in the output, or is it just the last one that's different?

@@moisesdepaulodias7980 I compared all the outputs of the earlier commands before the final one and all has exactly the same output in the video. I changed the runtime just like in the video. The last one only has the different output from the video. It doesn't show any of the v3 lines shown in the video. I can show the output in discord.

​@@ygg278 Hmm.. I just ran it here and I am not getting any error. How many images are you using? Also, are you sure that you uploaded the images AND the text files with the coordinates? (you can check the command before the "!python proccess.py" if the output contains one .txt after every .jpg Please call me on discord, my user is moises_dias, I will try my best to help you

@@moisesdepaulodias7980 I used 771 png images. i uploaded both images and text files with the coordinates.

@@moisesdepaulodias7980 I don't know why I can't see my reply but I'll just say it again, I used 771 png images. I added you already, daleggy.

opa, obrigado pelo elogio! pode me chamar no discord: moises_dias