Did you enjoy this video? Try my premium courses! 😃🙌😊 ● Hands-On Computer Vision in the Cloud: Building an AWS-based Real Time Number Plate Recognition System bit.ly/3RXrE1Y ● End-To-End Computer Vision: Build and Deploy a Video Summarization API bit.ly/3tyQX0M ● Computer Vision on Edge: Real Time Number Plate Recognition on an Edge Device bit.ly/4dYodA7 ● Machine Learning Entrepreneur: How to start your entrepreneurial journey as a freelancer and content creator bit.ly/4bFLeaC Learn to create AI-based prototypes in the Computer Vision School! www.computervision.school 😃🚀🎓
Hello from Mexico! I love your job, I did each step in the same way as you, and I had no difficulties, I really feel very grateful for the time you spent teaching us. Congratulations teacher! 👨🏫
For all who are getting errors like "inhomogeneous shapes" while training on big datasets take into account that the MP Hands processing not always return 42 features (sometimes it just doesn't predict the coordinates well enough). To avoid this situations always check the length of every array. You must have the same amount of images and labels, and the labels (landmark coordinates) should have the same shapes. Just remove the samples that doesn't return all the landmarks or doesn't work well with the Mediapipe hands solution, to ensure all the data has the same shape and to avoid these numpy errors (and bad models).
i think guys to solve this problem we had to tell the collect data script to save just frames where he could detect our hands else we will store bad models that will ends with this getting errors like "inhomogeneous shapes" , i actually try to solved this problem by not moving my hand when collecting data and making my model else you can try this code to check your images that are stored This script will only print the paths of the images that are deleted due to no hands being detected. It won't display any image windows. ##########################################" import os import cv2 import mediapipe as mp def process_and_show(image_path, mp_drawing): mp_hands = mp.solutions.hands hands = mp_hands.Hands()
# Detect hands and landmarks results = hands.process(image_rgb)
if not results.multi_hand_landmarks: print(f"Deleted image: {image_path}") # Delete the image with no hands detected os.remove(image_path) # Path to your data folder containing subfolders data_folder = "data" mp_drawing = mp.solutions.drawing_utils mp_drawing_styles = mp.solutions.drawing_styles # Iterate through subfolders for folder_name in os.listdir(data_folder): folder_path = os.path.join(data_folder, folder_name) if os.path.isdir(folder_path): print(f"Checking images in folder: {folder_name}") # Iterate through images in the folder for filename in os.listdir(folder_path): if filename.endswith(".jpg") or filename.endswith(".png"): image_path = os.path.join(folder_path, filename) process_and_show(image_path, mp_drawing)
Thank you so much, sir for this wonderful project. I've completed my term project easily with the help of your video. Loved how we can create our own data instead of getting it from somewhere else.
Sir!! You have my respect I have really learned lots of things in your whole video . Just keep making this ML/DL Project videos , that you have done like implementing from scratch any exciting ML/DL project. Just Keep Going Sir!!! Thankyou So much!!✨✨✨✨✨✨❤❤❤❤❤❤
hello from Nigeria i must say thanks for this video it was short, precise and educative yes, i had some errors which i was able to handle due to my past knowledge on Deep Learning. And for those that had issues with the disparity in the length of the data, you can always pad to its maximum length currently, i have a model that can identify 26 classes correctly and i will definitely increase the classes. i made each classes to have 700 images under different lighting condition thanks for all you do.
can you share your code? I'm having somre errors, while I try do identify the letters. Also, in your code, could you do with signs with both hands and with movements? @e2mnaturals442
For those who faces the error where it can't convert the 'data' values from dictionary data_dict, just make sure that in photo samples you are giving the full hand because if not, there will be inconsistent data and the lists will not have the same lenght inside the data_dict['data']. Do again the photos retrieve part and all should be fine
Really Thank you sir. Great Project you helped me a lot to learn many things. After multiple errors solving finally i succeeded in making full project.
You're great, Man,, thank you for teaching us and put lots of research first to ensure Windows user can replicate the project too,, . let me leave a logs here for other Windows users: 1. dont forget using packages with exactly same version as mentioned in requirements_windows.txt. 2. Use numpy 1.23.3 version,, I take a sneak peek to your terminal output that give me information if you use numpy with that version,, at first my terminal installed numpy 2.0 version, but no luck, and then dowgrade it,, 3. If you succesfully Instal Cmake via terminal, but still got error when compiling, I suggest you to install it by install Visual Studio first I've spent my first 4 hours dealing with those error before finally made it,,
Thank you, very clear what was taught. I want to ask what if the dataset from a public video had the initial and final movements? whether the start and end frames go into training . and using deep learning?
Hi! Great tutorial thank you. I have a question: does this program have data augmentation? and did u calculate the sensibility and accuracy of the program?
Hii!! I loved your video. I learned a lot. I just have one question, if at the end I want to form a sentence and print it, how can I save each character on the screen to have a full sentence at the end?
Thanks a lot! I really appreciate keeping this under an hour as well :)) We are trying to implement this model in Flutter to develop a mobile app. How can we create Flutter integration ?
Hello, I have watched your video and found it very informative. However, I was wondering if you could make a video for recognizing different characters for a sequence of movements, for example, the letter "J" or "Z." Thank you for your video.
Hello! Thank you so much for the tutorial!! :) Although I have trouble when trying to find the script's code at the very beginning, how can I get the code and connect my camera to get the 100 frames? Is it on GitHub? With what name? It seems to be there only the code that we built in the video...
Hola from India sir, Sir i enjoyed your video very much. sir, I have a small doubt can you tell me how to check and the accuracy of the model being trained.
great tutorial so helpful for my pfe project i actually have to do hand recognition identification biometric only but the hand contour you explained so well the part "this is the most important thing" and I really need help when it comes to the approach of how i can solve this if it? is possible for you to help me by doing a video of it ?cause its the first time for me working with python i usually work with Matlab. thank you again for this video
Hey Hayat, I am glad you found it helpful! 😄 Do you mean making a video about how to be strategic when starting a project and choose the most promising approach? Sure, I can do a video about problem solving strategies! 😃🙌
Hlo Sir, very nice video.... I also want to make a similar project ... But there will a bit difference.. I want to generate the entire subtitle for people who can't speak using their hand gestures during video conferencing in real time. Can you please guide me with the same ... Bcoz I completely a beginner. Your help will be appreciated. Thanks in advance. 😀
Hey Sourabh, it sounds like a complex and very cool project! I would start by saving all the symbols you detect, its confidence score, and the duration of time you detect them so you can analyze this info later on. This is going to help you to understand the problem a little better and also it is going to help you to define rules in order to achieve your goal. 😃💪
Hello , great tutorial 😀can this same approach be applied for british sign language because that uses both hands to make gestures , also can this be deployed in the real world and used at production level ?
You would need to make some edits in order to use it with both hands but I guess it would work, yes. Regarding the performance, yeah you could train it and improve it so it can be used at a production level. 🙌
Hey @@ComputerVisionEngineer , its not working efficiently for the british sign lang , maybe because it uses both hands , do you have any suggestions on how i can build up my project , it'll be a huge help , thanks
Do you mean using Yolo for object detection instead of mediapipe + Scikit learn? It can be done. You just need to train it. I did it with mediapipe + Scikit learn only for simplicity, and I think it also results in a more robust classifier. 🙌
I have trained my model using only numbers' data. It is working but the problem is it is only showing the numbers 9 or 1 in the frame. Do you think it's because of unclear data or problem in the training model. BTW great tutorial 👍
Hi... Since many signs involve some type of movement, I wonder if videos could be used in place of pictures. I hope you can reply to me because your video is very helpful for us. Thanks in advance.
I checked the github repo and there are some changes compared to the video. Why are you substracting the min of x_ from x (data_aux.append(x - min(x_))), also for y ? Why is it necessary to do that instead of just append x the way it is to the array. I saw u did that in the data processing and also in the model testing. Thanks a lot!
Hey George! Yeah, I sent that change in a new commit. It makes the solution more robust, you could think about it as a way of 'normalization'. This makes the classifier learn better than the (x, y) position of each landmark is not that important, the distance of each landmark to each other landmark is what matters most! 😃💪
@@ComputerVisionEngineer Thanks a lot for the answer! I thought it has something to do with the mediapipe library and is a must, but it actually makes sense to be some kind of normalization. Thanks for you time!
sir , the projects get closed if more hands are placed in the real-time video , i know that randomforest classifier uses only certain features , is there a way so that the program doesnt close if more hands are in the video
@@ComputerVisionEngineer Sir i have an error "ValueError: The least populated class in y has only 1 member, which is too few. The minimum number of groups for any class cannot be less than 2. ", what can be the problem, im trying to classfy all the alphabet letters, your help will be highly appreciated.
I am looking at the Indian sign language alphabet and I see some characters are done with 2 hands and others with 1 hand. In order to do something based on landmarks as we did on this video you would have to train 2 classifiers, one of them taking as input the landmarks of one hand only (as we did on the video) and the other classifier taking as input the landmarks of both hands. Then some logic to apply one classifier or the other one depending on how many hands appear on the frame. Or, you can just follow a different approach and train an image classifier taking the crop of the hand/s. 💪🙌
@@ocelottes it is mediapipe hand detection, if you want to test it's accuracy you would need to take another hand detector to compare mediapipe detections against
hello Sir, I follow your video for learning about computer vision . So I have a trouble with "DATA_DIR = './data'" , Is this file need to import from somewhere or should we need to prepare them? Can you help me to solve this?
First of all i want to thank you for this tutorial. I want actually to make a program for sign language but i am confused about the Dataset and how to process the Data which i will maybe get as Videos or Images. can you maybe give me some advice.
Hello, i was adding new alphabets to the dataset and got this error , unable to solve : " File "D:\Major project\.Major Project\code\train_classifier.py", line 11, in data = np.asarray(data_dict['data']) ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (400,) + inhomogeneous part."
I have tried it with arabic Sign language,and it did not working correctly, I get one letter almost every time and it's wrong letter, any ideas that can help me train the model. I got the dataset from kaggle.
EVERYTHING IS WORKING FINE, EXCEPT FOR THE FACT THAT THE MY FINAL PROGRAM IS UNABLE TO RECOGNIZE ANY SIGN. IT JUST GIVE EVERY SIGN THE SAME LABEL WHATEVER THERE IS IN THE INDEX 0 OF THE LABEL LIST. I don't understand why its not working???
did you use any particular research paper for this project. i have to make a report for my project and cite a reference and it would help if you can tell me which one you used or which one will be the most similar to this project.
@@ComputerVisionEngineer i think guys to solve this problem we had to tell the collect data script to save just frames where he could detect our hands else we will store bad models that will ends with this getting errors like "inhomogeneous shapes" , i actually try to solved this problem by not moving my hand when collecting data and making my model else you can try this code to check your images that are stored This script will only print the paths of the images that are deleted due to no hands being detected. It won't display any image windows. ##########################################" import os import cv2 import mediapipe as mp def process_and_show(image_path, mp_drawing): mp_hands = mp.solutions.hands hands = mp_hands.Hands()
# Detect hands and landmarks results = hands.process(image_rgb)
if not results.multi_hand_landmarks: print(f"Deleted image: {image_path}") # Delete the image with no hands detected os.remove(image_path) # Path to your data folder containing subfolders data_folder = "data" mp_drawing = mp.solutions.drawing_utils mp_drawing_styles = mp.solutions.drawing_styles # Iterate through subfolders for folder_name in os.listdir(data_folder): folder_path = os.path.join(data_folder, folder_name) if os.path.isdir(folder_path): print(f"Checking images in folder: {folder_name}") # Iterate through images in the folder for filename in os.listdir(folder_path): if filename.endswith(".jpg") or filename.endswith(".png"): image_path = os.path.join(folder_path, filename) process_and_show(image_path, mp_drawing)
Hi, while going through this code i'm getting model_dict = pickle.load(open('./model.p', 'rb')) FileNotFoundError: [Errno 2] No such file or directory: './model.p' and I didn't find any model.p file in your repository
I am getting plots for every data set size which i have taken is it fine bcs i have plt.savefig function, annotated it so that the plt for every dataset size is saved in main data directory
Hello sir, Kindly solve this error for me ----> ValueError: With n_samples=1, test_size=0.2 and train_size=0.8, the resulting train set will be empty. Adjust any of the aforementioned parameters.
How can I get accuracy for the letters predicted? Basically I want live accuracy for the letters that are predicted , since if you show any random hand gesture it will always predict some random letter, so it will be much better if you could also show live accuracy .Is it possible can u guide me a little bit through this?
Try using the method 'predict_proba' instead of 'predict'. You wil get a probability vector for all the classes. Taking the largest number will give you the confidence value you are looking for. 💪💪
@@prathamupadhyay1265 bhai if u dont mind kya app apke code ki zip file mujhe share kar skte hai, coz im getting many errors and i have tried many steps but kuch ho nahi raha hai. PLZ!!!!!!
Hi sir, i got some error at inference_classifier.py, the errors says: Line 36, in H, W, _= frame.shape AttributeError: ‘NoneType’ object has no attribute ‘shape’ Thank you for the help🙏🏻
i think guys to solve this problem we had to tell the collect data script to save just frames where he could detect our hands else we will store bad models that will ends with this getting errors like "inhomogeneous shapes" , i actually try to solved this problem by not moving my hand when collecting data and making my model else you can try this code to check your images that are stored This script will only print the paths of the images that are deleted due to no hands being detected. It won't display any image windows. ##########################################" import os import cv2 import mediapipe as mp def process_and_show(image_path, mp_drawing): mp_hands = mp.solutions.hands hands = mp_hands.Hands()
# Detect hands and landmarks results = hands.process(image_rgb)
if not results.multi_hand_landmarks: print(f"Deleted image: {image_path}") # Delete the image with no hands detected os.remove(image_path) # Path to your data folder containing subfolders data_folder = "data" mp_drawing = mp.solutions.drawing_utils mp_drawing_styles = mp.solutions.drawing_styles # Iterate through subfolders for folder_name in os.listdir(data_folder): folder_path = os.path.join(data_folder, folder_name) if os.path.isdir(folder_path): print(f"Checking images in folder: {folder_name}") # Iterate through images in the folder for filename in os.listdir(folder_path): if filename.endswith(".jpg") or filename.endswith(".png"): image_path = os.path.join(folder_path, filename) process_and_show(image_path, mp_drawing)
@ComputerVisionEngineer ValueError: X has 84 features, but RandomForestClassifier is expecting 42 features as input..I am getting this error when i run the inference_clasifier.py model...What change should i make in the code.....
If you're getting this, that means you're showing something else that isn't in the data. Only show what you've captured. Or else simply increase number of classes and take different pictures from different angles.
i think guys to solve this problem we had to tell the collect data script to save just frames where he could detect our hands else we will store bad models that will ends with this getting errors like "inhomogeneous shapes" , i actually try to solved this problem by not moving my hand when collecting data and making my model else you can try this code to check your images that are stored This script will only print the paths of the images that are deleted due to no hands being detected. It won't display any image windows. ##########################################" import os import cv2 import mediapipe as mp def process_and_show(image_path, mp_drawing): mp_hands = mp.solutions.hands hands = mp_hands.Hands()
# Detect hands and landmarks results = hands.process(image_rgb)
if not results.multi_hand_landmarks: print(f"Deleted image: {image_path}") # Delete the image with no hands detected os.remove(image_path) # Path to your data folder containing subfolders data_folder = "data" mp_drawing = mp.solutions.drawing_utils mp_drawing_styles = mp.solutions.drawing_styles # Iterate through subfolders for folder_name in os.listdir(data_folder): folder_path = os.path.join(data_folder, folder_name) if os.path.isdir(folder_path): print(f"Checking images in folder: {folder_name}") # Iterate through images in the folder for filename in os.listdir(folder_path): if filename.endswith(".jpg") or filename.endswith(".png"): image_path = os.path.join(folder_path, filename) process_and_show(image_path, mp_drawing)
@@ComputerVisionEngineer Hey, I mean the google coral yes. I want to connect it to my raspberrypi with a camera and try object recognition, especially for hand signs
@@prois5999 I need to take a look at it, the coral accelerator doesn't work with any architecture, so most likely will need to be something like mobilenet or efficiendet, instead of Yolo.
Hi sir, Thanks for your tutorial. Yet, I a problem in locating the file(./data), and received an error message of [Errno 20] Not a directory: './data/.DS_Store'. while using "create_dataset.py". Currently all file are put in desktop, do you know why? (I m using MacBook)
The thing about Apple is that MacOS often puts a file called ".DS_Store" in the directory which stores some information. In your code where you iterate over folders, compare the name with ".DS_Store" and simply skip it
Hello sir, I got a one problem. I made the same with you and my code is worked but it only showed at least 5 mins for capturing then the camera will shutdown automatically and got some errors. :((((
Sir please help............during training it shows value error..data = np.asarray(data_dict['data']) ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (199,) + inhomogeneous part......for 3 class
Did you enjoy this video? Try my premium courses! 😃🙌😊
● Hands-On Computer Vision in the Cloud: Building an AWS-based Real Time Number Plate Recognition System bit.ly/3RXrE1Y
● End-To-End Computer Vision: Build and Deploy a Video Summarization API bit.ly/3tyQX0M
● Computer Vision on Edge: Real Time Number Plate Recognition on an Edge Device bit.ly/4dYodA7
● Machine Learning Entrepreneur: How to start your entrepreneurial journey as a freelancer and content creator bit.ly/4bFLeaC
Learn to create AI-based prototypes in the Computer Vision School! www.computervision.school 😃🚀🎓
Hello from Mexico!
I love your job, I did each step in the same way as you, and I had no difficulties, I really feel very grateful for the time you spent teaching us.
Congratulations teacher!
👨🏫
Thank you! So glad you enjoy the content! 😃🙌
Could you tell me the installation process
For all who are getting errors like "inhomogeneous shapes" while training on big datasets take into account that the MP Hands processing not always return 42 features (sometimes it just doesn't predict the coordinates well enough).
To avoid this situations always check the length of every array. You must have the same amount of images and labels, and the labels (landmark coordinates) should have the same shapes.
Just remove the samples that doesn't return all the landmarks or doesn't work well with the Mediapipe hands solution, to ensure all the data has the same shape and to avoid these numpy errors (and bad models).
can you help me. when I trained only one gesture nothing else, but the system detects untrained gestures as the trained gesture why? any idea
can you please share the changed code
i think guys to solve this problem we had to tell the collect data script to save just frames where he could detect our hands else we will store bad models that will ends with this getting errors like "inhomogeneous shapes" , i actually try to solved this problem by not moving my hand when collecting data and making my model else you can try this code to check your images that are stored
This script will only print the paths of the images that are deleted due to no hands being detected. It won't display any image windows.
##########################################"
import os
import cv2
import mediapipe as mp
def process_and_show(image_path, mp_drawing):
mp_hands = mp.solutions.hands
hands = mp_hands.Hands()
# Read the image
image = cv2.imread(image_path)
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# Detect hands and landmarks
results = hands.process(image_rgb)
if not results.multi_hand_landmarks:
print(f"Deleted image: {image_path}")
# Delete the image with no hands detected
os.remove(image_path)
# Path to your data folder containing subfolders
data_folder = "data"
mp_drawing = mp.solutions.drawing_utils
mp_drawing_styles = mp.solutions.drawing_styles
# Iterate through subfolders
for folder_name in os.listdir(data_folder):
folder_path = os.path.join(data_folder, folder_name)
if os.path.isdir(folder_path):
print(f"Checking images in folder: {folder_name}")
# Iterate through images in the folder
for filename in os.listdir(folder_path):
if filename.endswith(".jpg") or filename.endswith(".png"):
image_path = os.path.join(folder_path, filename)
process_and_show(image_path, mp_drawing)
I understood the problem but can't really put it in the program. could you explain it please?
Just add in create_dataset :
if (len(data_aux) == 42):
data.append(data_aux)
labels.append(dir_)
Thank you so much, sir for this wonderful project. I've completed my term project easily with the help of your video. Loved how we can create our own data instead of getting it from somewhere else.
Can u pls help me out? Please
Sir!! You have my respect I have really learned lots of things in your whole video . Just keep making this ML/DL Project videos , that you have done like implementing from scratch any exciting ML/DL project.
Just Keep Going Sir!!!
Thankyou So much!!✨✨✨✨✨✨❤❤❤❤❤❤
Thank you for your support! 😃🙌💪
great tutorial on how to organize the project into separate steps!
Good organization is the key to a successful project I am happy you enjoyed the video! 😄🙌
Having trouble with my ML project now, but so happy to find your video. Thanks for all the work!!
hello from Nigeria
i must say thanks for this video
it was short, precise and educative
yes, i had some errors which i was able to handle due to my past knowledge on Deep Learning. And for those that had issues with the disparity in the length of the data, you can always pad to its maximum length
currently, i have a model that can identify 26 classes correctly and i will definitely increase the classes. i made each classes to have 700 images under different lighting condition
thanks for all you do.
bro can you send me the file for your project
@@ijaspr5486 like the whole file?
Could you share your GitHub link of your project?
@@e2mnaturals442 yes like github code or i give you my social media id
can you share your code? I'm having somre errors, while I try do identify the letters. Also, in your code, could you do with signs with both hands and with movements? @e2mnaturals442
For those who faces the error where it can't convert the 'data' values from dictionary data_dict, just make sure that in photo samples you are giving the full hand because if not, there will be inconsistent data and the lists will not have the same lenght inside the data_dict['data']. Do again the photos retrieve part and all should be fine
Thanks a lot!! How did you notice that this was the issue?
thanks a lot bro!!!
It worked ! Thanks
Really Thank you sir. Great Project you helped me a lot to learn many things. After multiple errors solving finally i succeeded in making full project.
Glad the content is helpful! 😃🙌
srsly like the best video, now i can train my custom hand gestures etc. even, thank youu❤❤
Thanks a lot bro, I watched many videos and i wasted a lot of time and finally found your video and done my project.
You are welcome! Glad it was helpful! 😃
Please send your github link please
I got lot of error bro please please please please
You're great, Man,, thank you for teaching us and put lots of research first to ensure Windows user can replicate the project too,,
.
let me leave a logs here for other Windows users:
1. dont forget using packages with exactly same version as mentioned in requirements_windows.txt.
2. Use numpy 1.23.3 version,, I take a sneak peek to your terminal output that give me information if you use numpy with that version,, at first my terminal installed numpy 2.0 version, but no luck, and then dowgrade it,,
3. If you succesfully Instal Cmake via terminal, but still got error when compiling, I suggest you to install it by install Visual Studio first
I've spent my first 4 hours dealing with those error before finally made it,,
Thanks a lot! I really appreciate keeping this under an hour as well :))
The best tutorial ever!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
very awesome tutorial with brilliant idea and conceptualization. Thanks a lost Felipe!
Thank you for your support! Glad you enjoyed it! 😃🙌
Thank you, very clear what was taught. I want to ask what if the dataset from a public video had the initial and final movements? whether the start and end frames go into training . and using deep learning?
i love all that, you are very clearly and simply 😍😍
Thank you! Glad you enjoyed it! 😃💪
Hi! Great tutorial thank you. I have a question: does this program have data augmentation? and did u calculate the sensibility and accuracy of the program?
Hii!! I loved your video. I learned a lot. I just have one question, if at the end I want to form a sentence and print it, how can I save each character on the screen to have a full sentence at the end?
Hi, did you get it?
Thanks a lot! I really appreciate keeping this under an hour as well :)) We are trying to implement this model in Flutter to develop a mobile app. How can we create Flutter integration ?
Hello, I have watched your video and found it very informative. However, I was wondering if you could make a video for recognizing different characters for a sequence of movements, for example, the letter "J" or "Z." Thank you for your video.
I will try to. 🙌
Mas argentino imposible jsjs, Gran video!
Hello! Thank you so much for the tutorial!! :)
Although I have trouble when trying to find the script's code at the very beginning, how can I get the code and connect my camera to get the 100 frames? Is it on GitHub? With what name? It seems to be there only the code that we built in the video...
Hola from India sir, Sir i enjoyed your video very much. sir, I have a small doubt can you tell me how to check and the accuracy of the model being trained.
great tutorial so helpful for my pfe project i actually have to do hand recognition identification biometric only but the hand contour you explained so well the part "this is the most important thing" and I really need help when it comes to the approach of how i can solve this if it? is possible for you to help me by doing a video of it ?cause its the first time for me working with python i usually work with Matlab. thank you again for this video
Hey Hayat, I am glad you found it helpful! 😄 Do you mean making a video about how to be strategic when starting a project and choose the most promising approach? Sure, I can do a video about problem solving strategies! 😃🙌
Do you have file word report ?
thank you my teacher, great a video , i tried it myself, I did it :)
You are welcome! 😃 Glad you enjoyed it!! 🙂🙌
Hlo Sir, very nice video.... I also want to make a similar project ... But there will a bit difference.. I want to generate the entire subtitle for people who can't speak using their hand gestures during video conferencing in real time.
Can you please guide me with the same ... Bcoz I completely a beginner. Your help will be appreciated. Thanks in advance. 😀
Hey Sourabh, it sounds like a complex and very cool project! I would start by saving all the symbols you detect, its confidence score, and the duration of time you detect them so you can analyze this info later on. This is going to help you to understand the problem a little better and also it is going to help you to define rules in order to achieve your goal. 😃💪
Hi Sourabh were you able to make it if yes could you please share some update or code for the same
Hello , great tutorial 😀can this same approach be applied for british sign language because that uses both hands to make gestures , also can this be deployed in the real world and used at production level ?
You would need to make some edits in order to use it with both hands but I guess it would work, yes. Regarding the performance, yeah you could train it and improve it so it can be used at a production level. 🙌
thanks @@ComputerVisionEngineer 😁i'll try and see if it works out
Hey @@ComputerVisionEngineer , its not working efficiently for the british sign lang , maybe because it uses both hands , do you have any suggestions on how i can build up my project , it'll be a huge help , thanks
Thank you so much it's helpful for me 😊
Glad to hear it is helpful! 😃🙌
size.width>0 && size.height>0 in function 'cv::imshow' error sir
Really great video tutorial! Why did you choose scikt learn and not Yolo? How many changes would you have to make to use Yolo?
Do you mean using Yolo for object detection instead of mediapipe + Scikit learn? It can be done. You just need to train it. I did it with mediapipe + Scikit learn only for simplicity, and I think it also results in a more robust classifier. 🙌
Thanks for your good tutorial
How to act for the rest of the letters?
I have trained my model using only numbers' data. It is working but the problem is it is only showing the numbers 9 or 1 in the frame. Do you think it's because of unclear data or problem in the training model.
BTW great tutorial 👍
Thank you sir for this video
Hi... Since many signs involve some type of movement, I wonder if videos could be used in place of pictures. I hope you can reply to me because your video is very helpful for us. Thanks in advance.
Yes, you could try with video classification. 🙌
@@ComputerVisionEngineer how to insert video type in pycharm?
I hope you can help us..thank you
Is there a front - end that can connect in pycharm?
Great video but How do you do the collecting images part of the code?
I checked the github repo and there are some changes compared to the video. Why are you substracting the min of x_ from x (data_aux.append(x - min(x_))), also for y ? Why is it necessary to do that instead of just append x the way it is to the array. I saw u did that in the data processing and also in the model testing. Thanks a lot!
Hey George! Yeah, I sent that change in a new commit. It makes the solution more robust, you could think about it as a way of 'normalization'. This makes the classifier learn better than the (x, y) position of each landmark is not that important, the distance of each landmark to each other landmark is what matters most! 😃💪
@@ComputerVisionEngineer Thanks a lot for the answer! I thought it has something to do with the mediapipe library and is a must, but it actually makes sense to be some kind of normalization. Thanks for you time!
Great content, thank you so much.
You are welcome!! 😃
Just wanted to tell you that your project is very famous in SMIT 😊
😃 That is soooo cool! I am happy to help you guys. 😊🙌
Have you manage to Train the model with all alphabet letters
sir , the projects get closed if more hands are placed in the real-time video , i know that randomforest classifier uses only certain features , is there a way so that the program doesnt close if more hands are in the video
hehe subscribed, tysm for this it was very helpful
why does it close when you put another hand?
Hello, I have watched your video and found it very informative. However, I was wondering what is the limitation of this project?
Hey, limitation in terms of possible symbols? I would say any static symbol made with only one hand.
Some hand sign have two hand ,than what we can do that situation ?
another great project
Thank you, Arif! I am happy you enjoyed it. 😃🙌
Sir, if i would create 10 samples different of course what should i do at the stage of using the source code? please give me an example🙏🏻
Great Video
why do you use and random forest classifier algorithm?
maybe it is better for it?
could i try with a pretrained model to get better results?
No particular reason why I used a Random Forest, I think pretty much any other classifier would have a similar performance in this case.
@@ComputerVisionEngineer Thanks felipe!!
Sir only 9 character can be trained plz help me to train 26 character
amazing project, i want to do it but with raspberry pi, some suggestion?
Thank you! I haven't tried to do it an edge device, I don't have any suggestions. 🙌
I have just subscribed,
Currently working on a similar project, fingers crossed I'm at a right place..😂
🤞😀 Good luck with your project, Martin! 🙌
@@ComputerVisionEngineer Sir i have an error "ValueError: The least populated class in y has only 1 member, which is too few. The minimum number of groups for any class cannot
be less than 2.
", what can be the problem, im trying to classfy all the alphabet letters, your help will be highly appreciated.
@@martinsilungwe2725 do you have any solution for it now?
I am new to AI. I just want to know are we using Natural Language, Machine Learning and computer vision.
Great tutorial! Can you tell me how can I do this for Indian Sign Language which uses 2 hands?
I am looking at the Indian sign language alphabet and I see some characters are done with 2 hands and others with 1 hand. In order to do something based on landmarks as we did on this video you would have to train 2 classifiers, one of them taking as input the landmarks of one hand only (as we did on the video) and the other classifier taking as input the landmarks of both hands. Then some logic to apply one classifier or the other one depending on how many hands appear on the frame. Or, you can just follow a different approach and train an image classifier taking the crop of the hand/s. 💪🙌
Hi ! Have you completed working on this project? Did it worked ?
Very cool, i have a question. How can i test de accuracy of the detection?
Do you mean the accuracy of the hand detection?
@@ComputerVisionEngineer yes
@@ocelottes it is mediapipe hand detection, if you want to test it's accuracy you would need to take another hand detector to compare mediapipe detections against
great project! may i ask what algorithm is used in your sign language?
Hey, thank you! I am using mediapipe as a hand detector and landmark detector and a Random Forest classifier as sign classifier. 🙌
hello Sir, I follow your video for learning about computer vision .
So I have a trouble with "DATA_DIR = './data'" , Is this file need to import from somewhere or should we need to prepare them? Can you help me to solve this?
am also thinking the same. The images seem no to be there
can we make this project with pose detection models like openpose or deeppose? and what is the difference
First of all i want to thank you for this tutorial. I want actually to make a program for sign language but i am confused about the Dataset and how to process the Data which i will maybe get as Videos or Images. can you maybe give me some advice.
Sure. Do you think you can take an approach as I do in the video?
The camera crashes when I show more than one hand. Can you tell me how it can be fixed?
Thank you for the video, can you also make a video on sign language recognition on a video dataset (Word level american sign language dataset).
You are welcome! I will try to make a video about it. 🙌
Hello! can you please tell me which ML algorithm you used here???
life saver.
will this work the same if i wanna use two hands??
because in the indian sign language we use two hands
Hello, i was adding new alphabets to the dataset and got this error , unable to solve : " File "D:\Major project\.Major Project\code\train_classifier.py", line 11, in
data = np.asarray(data_dict['data'])
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (400,) + inhomogeneous part."
did you fix this error?
how do i get that function 18:10 ?
which software is used to execute this process
thanks
Hello!! Can you tell me which ML algorithm did you use in this?
I have tried it with arabic Sign language,and it did not working correctly, I get one letter almost every time and it's wrong letter, any ideas that can help me train the model. I got the dataset from kaggle.
EVERYTHING IS WORKING FINE, EXCEPT FOR THE FACT THAT THE MY FINAL PROGRAM IS UNABLE TO RECOGNIZE ANY SIGN. IT JUST GIVE EVERY SIGN THE SAME LABEL WHATEVER THERE IS IN THE INDEX 0 OF THE LABEL LIST. I don't understand why its not working???
same here, did you fix it?
The app crashes when using both hands. How can I fix this?
Thanks man
did you use any particular research paper for this project. i have to make a report for my project and cite a reference and it would help if you can tell me which one you used or which one will be the most similar to this project.
Hey, I didn't use any research paper for this project. 🙌
@@ComputerVisionEngineer alright then...but do you have any idea which one would be similar or near to this?
@@054_vishwadhimar4 hi, did you get the research paper?
@@aakritityagi7203 no I did not actually... thankfully my mentor did not force me to find one and accepted multiple youtube.videos as references
If you train it in a specific place ex: your bedroom would this work like with the background of your kitchen or different place?
Yes, by the way we are doing it in this tutorial, it should work if you change the background. 🙌
@@ComputerVisionEngineer i think guys to solve this problem we had to tell the collect data script to save just frames where he could detect our hands else we will store bad models that will ends with this getting errors like "inhomogeneous shapes" , i actually try to solved this problem by not moving my hand when collecting data and making my model else you can try this code to check your images that are stored
This script will only print the paths of the images that are deleted due to no hands being detected. It won't display any image windows.
##########################################"
import os
import cv2
import mediapipe as mp
def process_and_show(image_path, mp_drawing):
mp_hands = mp.solutions.hands
hands = mp_hands.Hands()
# Read the image
image = cv2.imread(image_path)
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# Detect hands and landmarks
results = hands.process(image_rgb)
if not results.multi_hand_landmarks:
print(f"Deleted image: {image_path}")
# Delete the image with no hands detected
os.remove(image_path)
# Path to your data folder containing subfolders
data_folder = "data"
mp_drawing = mp.solutions.drawing_utils
mp_drawing_styles = mp.solutions.drawing_styles
# Iterate through subfolders
for folder_name in os.listdir(data_folder):
folder_path = os.path.join(data_folder, folder_name)
if os.path.isdir(folder_path):
print(f"Checking images in folder: {folder_name}")
# Iterate through images in the folder
for filename in os.listdir(folder_path):
if filename.endswith(".jpg") or filename.endswith(".png"):
image_path = os.path.join(folder_path, filename)
process_and_show(image_path, mp_drawing)
The mediapipe library is giving error in installation what should I do?
did you figure it out?
My problem is no such file or directory ; /data. Pickle
Hi, while going through this code i'm getting model_dict = pickle.load(open('./model.p', 'rb'))
FileNotFoundError: [Errno 2] No such file or directory: './model.p' and I didn't find any model.p file in your repository
Hey, you can create the model yourself following the steps I describe in the video. 😃🙌
I am getting plots for every data set size which i have taken is it fine bcs i have plt.savefig function, annotated it so that the plt for every dataset size is saved in main data directory
Hello sir, Kindly solve this error for me ----> ValueError: With n_samples=1, test_size=0.2 and train_size=0.8, the resulting train set will be empty. Adjust any of the aforementioned parameters.
Hey, would you please copy paste the full description of the error you get?
Sir what is your python version
can i ask how can you moved this into mobile / android studio
Is the program you are using called pycharm?
Yes. 🙌
How can I get accuracy for the letters predicted?
Basically I want live accuracy for the letters that are predicted , since if you show any random hand gesture it will always predict some random letter, so it will be much better if you could also show live accuracy .Is it possible can u guide me a little bit through this?
Try using the method 'predict_proba' instead of 'predict'. You wil get a probability vector for all the classes. Taking the largest number will give you the confidence value you are looking for. 💪💪
@@ComputerVisionEngineer Thanks a lot you are amazing !!! 😃
@@prathamupadhyay1265 bhai if u dont mind kya app apke code ki zip file mujhe share kar skte hai, coz im getting many errors and i have tried many steps but kuch ho nahi raha hai. PLZ!!!!!!
plz bhai
@@yashanchule9641 GitHub link is there..or have you tried that too?!
how to make this project on web based like on react or flask
Hi sir, i got some error at inference_classifier.py, the errors says:
Line 36, in
H, W, _= frame.shape
AttributeError: ‘NoneType’ object has no attribute ‘shape’
Thank you for the help🙏🏻
It's fully working for you now?
Because I cannot able to run the first step please help mee
In collect_img is cv2.imshow(frame) is error bro kindly help me
Error name :size.width>0 && size.height>0 in function 'cv::imshow'
can you please show the err
@@RohanVector
Change the line to -> cap = cv2.VideoCapture(0)...
Previously it was -> cap = cv2.VideoCapture(2)@@RohanVector
Who can I add more sign because it's getting error when I try to add more signs
it's showing the error: ValueError: setting an array element with a sequence.
after loading the dictionary in the model.
i think guys to solve this problem we had to tell the collect data script to save just frames where he could detect our hands else we will store bad models that will ends with this getting errors like "inhomogeneous shapes" , i actually try to solved this problem by not moving my hand when collecting data and making my model else you can try this code to check your images that are stored
This script will only print the paths of the images that are deleted due to no hands being detected. It won't display any image windows.
##########################################"
import os
import cv2
import mediapipe as mp
def process_and_show(image_path, mp_drawing):
mp_hands = mp.solutions.hands
hands = mp_hands.Hands()
# Read the image
image = cv2.imread(image_path)
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# Detect hands and landmarks
results = hands.process(image_rgb)
if not results.multi_hand_landmarks:
print(f"Deleted image: {image_path}")
# Delete the image with no hands detected
os.remove(image_path)
# Path to your data folder containing subfolders
data_folder = "data"
mp_drawing = mp.solutions.drawing_utils
mp_drawing_styles = mp.solutions.drawing_styles
# Iterate through subfolders
for folder_name in os.listdir(data_folder):
folder_path = os.path.join(data_folder, folder_name)
if os.path.isdir(folder_path):
print(f"Checking images in folder: {folder_name}")
# Iterate through images in the folder
for filename in os.listdir(folder_path):
if filename.endswith(".jpg") or filename.endswith(".png"):
image_path = os.path.join(folder_path, filename)
process_and_show(image_path, mp_drawing)
Hi,
I am getting an error that ./data/.DS_Store is not a directory and is not found.
Hey, what file / line triggers this error?
chilean spanish?? good tutorial
Hi, thank you! My home country is Uruguay. 🙌
@ComputerVisionEngineer ValueError: X has 84 features, but RandomForestClassifier is expecting 42 features as input..I am getting this error when i run the inference_clasifier.py model...What change should i make in the code.....
If you're getting this, that means you're showing something else that isn't in the data. Only show what you've captured. Or else simply increase number of classes and take different pictures from different angles.
i think guys to solve this problem we had to tell the collect data script to save just frames where he could detect our hands else we will store bad models that will ends with this getting errors like "inhomogeneous shapes" , i actually try to solved this problem by not moving my hand when collecting data and making my model else you can try this code to check your images that are stored
This script will only print the paths of the images that are deleted due to no hands being detected. It won't display any image windows.
##########################################"
import os
import cv2
import mediapipe as mp
def process_and_show(image_path, mp_drawing):
mp_hands = mp.solutions.hands
hands = mp_hands.Hands()
# Read the image
image = cv2.imread(image_path)
image_rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
# Detect hands and landmarks
results = hands.process(image_rgb)
if not results.multi_hand_landmarks:
print(f"Deleted image: {image_path}")
# Delete the image with no hands detected
os.remove(image_path)
# Path to your data folder containing subfolders
data_folder = "data"
mp_drawing = mp.solutions.drawing_utils
mp_drawing_styles = mp.solutions.drawing_styles
# Iterate through subfolders
for folder_name in os.listdir(data_folder):
folder_path = os.path.join(data_folder, folder_name)
if os.path.isdir(folder_path):
print(f"Checking images in folder: {folder_name}")
# Iterate through images in the folder
for filename in os.listdir(folder_path):
if filename.endswith(".jpg") or filename.endswith(".png"):
image_path = os.path.join(folder_path, filename)
process_and_show(image_path, mp_drawing)
Do not give 2 hands at the same on your camera
Am from Ghana
hi I am a 15 year old and i want to do this for my school tech convention. What program are you using to code this
hey its does not work for more than 5 sign can show value error about the shape can you please fix it
How will this accept video feed from a phone ?
what si the data.pickle?
Could also do this with YOLO, OpenCV and the Coral?
Do you mean the coral accelerator?
@@ComputerVisionEngineer Hey, I mean the google coral yes. I want to connect it to my raspberrypi with a camera and try object recognition, especially for hand signs
Do you think this would work with your tutorial?
@@prois5999 I need to take a look at it, the coral accelerator doesn't work with any architecture, so most likely will need to be something like mobilenet or efficiendet, instead of Yolo.
Hi sir,
Thanks for your tutorial.
Yet, I a problem in locating the file(./data), and received an error message of [Errno 20] Not a directory: './data/.DS_Store'. while using "create_dataset.py". Currently all file are put in desktop, do you know why? (I m using MacBook)
The thing about Apple is that MacOS often puts a file called ".DS_Store" in the directory which stores some information. In your code where you iterate over folders, compare the name with ".DS_Store" and simply skip it
Hello sir, I got a one problem. I made the same with you and my code is worked but it only showed at least 5 mins for capturing then the camera will shutdown automatically and got some errors. :((((
Sir please help............during training it shows value error..data = np.asarray(data_dict['data'])
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (199,) + inhomogeneous part......for 3 class
i got the same error, were you able to solve it?
Did u solve it?
i was able to sort it using padding
if you want me to explain more, i will be glad to
@@e2mnaturals442 please explain
@@e2mnaturals442can u please explain it
doesnt this work when you use labeled dataset