Real Time Sign Language Detection with Tensorflow Object Detection and Python | Deep Learning SSD 

For those who came here after seeing the Indian girl's post, yeah Priyanjali Gupta plagiarised this repo 😂

The most logical and understandable guide on the subject I have seen. Great job.

You are my Prof in python. Thank you so much for all, the codes you share, the video and the tutorial you gave to us. 👍

dude. you. are. sick. thank you man. keep pushing. you could really become something big. super high quality content. cheers

Awesome tutorial. I am really glad that I found you!

I was looking for helpful vids on my project and this guy already did my project. Nice

Nick - Love your channel - Thanks for this video. I was still struggling after the "face mask" video. This one clarified a lot of the important points. It gave me a much better understanding of what is going on than just watching the first object detection video twice... ok ten times ;-) . Speed of training - I have an AMD RX 580 video card instead of NVidia so I'm stuck with training via my AMD 3700X CPU (or am I?). No hindrance to the learning experience of course, but if I had only known last year when I built up my systems... sigh. Thanks again, you are truly a terrific teacher!

This is really impressive! Can't wait to try it out.

this guy is awesome, hope to work with you one day Nicholas

Dear Nich Sir, I was a software engineer student, yesterday my results has come out and I got a first-class honor for my BSc degree, For the final project, I got 80 marks. My final project was Sign language detection in my language (Sinhala language). That project was kinda Advanced, but your sign language detection videos series inspired me to make a better project for my finals. So Thank you soo much and wanna say this, again and again, YOU ARE THE BEST AI TEACHER IN RU-vid, Keep making these kinda videos and make others' life brighter like this. I think you should pin this comment. ⭐⭐⭐

I'm loving the contents of this channel!

I finally completed this project Thanks Nick for your wonderful explanation and super simple guide lines 👍 Amazing video

Wouaouhhhhh... Man, your amazing :) ... This is what I was searching for after some months of Machine Learing study

Hey one girl has copied your code 😂 she is getting more fame than this video literally she is 😂😂😂😂😂😂😂😂😂😂😂

Excellent explanation... Thanks for sharing!

Nick, you are amazing! All thanks to you for making a well organised video with soo much clarity 💥💥 .... I did my first deep learning project based on your teaching and safe to say that it was a HIT among my professors.... I look forward to make more of such projects... Thank you so much Prof. Nick .... You are the BEST⭐⭐⭐⭐⭐

Hey y'all! Wrote up a couple of steps to setup your environment. I'm planning on doing a video on this but until then, here's a crash course post: 1. First up install Python 3.7.4. Download and install the package for your OS that has the words 2019.07 in it from here repo.anaconda.com/archive/. This should give you 3.7.4 to work with. 2. Then install Visual Studio C++ 2015 from here: go.microsoft.com/fwlink/?LinkId=691126. Tensorflow needs this in order to compile 3. OPTIONAL IF YOU HAVE A GPU - Install Cuda and Cudnn. Install Cuda first, then install Cudnn. - Cuda: 10.1 - developer.nvidia.com/cuda-10.1-download-archive-base - Cudnn: 7.6.5 - developer.nvidia.com/rdp/cudnn-download Once Cudnn is installed you need to copy the Cudnn files into their respective folders inside the Cuda directory. I used this as a guide: towardsdatascience.com/installing-tensorflow-with-cuda-cudnn-and-gpu-support-on-windows-10-60693e46e781 4. Then install Protoc 3.13 from here: github.com/protocolbuffers/protobuf/releases. For windows, download the repository and then add it to your PATH file. 5. Then install python packages using the pip command: Tensorflow: 2.3.1 - pip install tensorflow==2.3.1 OpenCV: 4.4.0 - pip install tensorflow==4.4.0 6. Install the object detection API. To do this, run these commands from a command prompt or terminal: git clone github.com/tensorflow/models cd tensorflow/models/research protoc object_detection/protos/*.proto --python_out=. cp object_detection/packages/tf2/setup.py . python -m pip install .

How do you not have more views? This is amazing content!

Just in time, I'm developing a project for my master's degree about this topic, thanks for the content

Really good explanation.. and real time experience is awesome.. thank a lot.. 👏

This is lit 🔥 - keep it coming!

Finally a video that makes sense !!!

Dear Nicholas, I would first like to thank you for showing us these amazing tutorials, they are nothing short of educational and fun. I came here asking for help though, if it is ok with you. I am developing a similar project to yours where the Object detector would figure out which technique in Judo it is observing. I am trying to run the project and I followed the tutorial to a T. I am able to train the model, but the loss value always starts at NaN (technically I am not able to train the model, but step 6 is within reach, because I can execute the command). I have gone through the comments and have done all the suggestions, which include: 1. Double checking the labels 2. Decrease the learning rate (although that is only viable if the loss value does not start at NaN) 3. Used 2 different datasets (1st is the Judo dataset, the other is the image dataset that was created, the same as yours). 4. Followed the installation to everything on your previous videos (Tensorflow installation + CUDA/CUDNN and Face mask detection tutorial) It would be greatly appreciated if you do end up helping me. But if not, this is still a very interesting video.

Your videos are so encouraging... 🙌🙌🙌

This is amazing Great content....Really helped me with one of the projects I was working on Keep making videos like these

This tutorial is great. Love the explanation

Thank you so much for this wonderful tutorial

Very useful video. Thankyou so much .

Are you awesome. Greetings from Colombia.

Thanks for this amazing tutorial Nick. You explained everything so nicely. A quick question, if I trained this model previously for 20 pictures of each class, and now I wish to increase the training dataset to around 50 pics. Will I have to retrain the model from scratch? (Cause on CPU, it took me around 5 hours to train for 20k steps!)

Nicholas! I have solved all errors. Although it was not easy, all codes worked well . Thank you so much for upload this tutorial!

had a great learning though i got a lots of errors in the way but that make me go deep into subject and taking this from scratch is awesome and again expectinga a lot like these from your side nick

Thank you so much for this tutorial Nicholas ! One more thing I wanted to ask that is it possible to read out the phrases that gets recognized by the system? If yes, please do guide !

Great video. Appreciate your effort a lot! Two questions: If you would perform two gestures using both hands at the same time, would the model have detected both gestures at the same time? Is it possible to train for one class only? Best regards

Awesome tutorial!

Thanks for this tutorial. So, really helpful.

Hi Nicolas, finally I have been able to run everything. Although my recognition is not that accurate . I guess it would be nice to automate the part of selecting the gesture in every picture 🤣

Who are here after that LinkedIn post ? 😂

awesome!!!!! too happy to learn

Amazing work Nicholas

Thanks for this amazing tutorial. I used it to train a model exactly as you did. Do you know what the best way to load the model I have trained into the C++ API of tensorflow? Thank you again!

Hi I'm glad I found your video! First of all, You are awesome man! Keep it up! I'm planning to do a similar project as a research proposal for a graduate school application and I was wondering how can I train my model if the sign language uses hand movements rather than still poses.

That was really well explained, thank you. What I'm confused about is how do you make a Pass/Fail object detection system? I can see how you'd spot say a normal widget on a conveyor belt, but how would you detect a broken widget, or a misplaced cat on the conveyor belt?

Hello @NicholasRenotte Sir, Thanks for the amazing tutorial! I am glad that I found this savior tutorial last minute....Thankyou so much :)

Hello Nicholas, I know you made this project long ago, but still I would really be glad if you let know what are the algorithms used here. Can we use other algorithms to this for its better accuracy and results like KNN, logistic regression or decision tree, if yes, please guide how. And last, how to generate the classification report here. Please help as I`m really new even to Python language as I have been seasoned with JAVA by far.

thank you for this great video .I just have one question. this code just recognize every sign that we give to it. if we want to have a code with all sign to recognize every sentences and words , what should we do?

great JOB man, I'd like to thank you it's very nice video

thx u so much for taking this tutorial video. I just done my program from packages installation to real-time detection. but the I think i need to change another cnn/rcnn model and train it again. because my accuracy is wired lol.

Hi Nicholas. Thank you. Do you also have the final model that you trained with your images? I'd like to convert it to tflite and try it.

Hi Nicholas. Thank you for this amazing video! How long did it take you to train this model? I'm going to train a model for detection about 10-15 gestures, but I also want to make train dataset for each gesture slightly bigger. If you remember the approximate amount of time I think I can roughly estimate how long the training session will take for me.

Nick sir...this session was really helpful.. keep on rocking with coding🎉

Good stuff dude. I'm and embedded software guy trying to learn some of this stuff to control my TV with sign language. Haha

What version of python, tensorflow and opencv are you using?

Excuse me, I want to ask. At 11:01 minutes, in the program to take pictures automatically, my program has no errors at all, looks good. The camera is detected successfully, the frame appears, and the frame can display the image from the camera. After the program finishes (camera off) a new directory appears, but none of the images are saved. What happened? May I know the explanation? Thank you before!

What a great video!

Really like your smile, Nicholas👍

Thank you so much, sir! I wonder what if I train about 100 different signs and gestures? Should I alter the layers in that config file? How to deal with overfitting and underfitting problem?

Thank you so much for this content, you saved me with my senior project- which is ASL alphabet recognition (turns out MATLAB is terrible for someone new to this field)... however, how can we implement this style detector for letters like "j" and "z" which require movement? I have seen many papers about combining lstm's with cnn's but it seems like everyone is keeping their cards close to their chest and their are no actual good tutorials on this matter... what are the chances of you doing one of these within the next month or so lol?

I was looking for helpful vids on my project ,I would like to know how this project is different from other existing ones and how we can improve when compared to the other projects. Thanks

i really apreciate your help with dowloanding this software

Thank You so much for explaining the project code by code. I am facing a small issue. My TF records are not getting created. Do you have any idea why? I've gone through many videos of yours to understand where am I going wrong.

Hey Nicholas, in my first run I trained my model to 12000 steps. There is still a relatively high loss (0.25). How can I alter the training command to retrain starting from my latest checkpoint?

it's working thank you soo much for your information

Hi Nick great fan of your work. Can you do a video on temporal action localisation on an untrimmed videos. I saw a lot of interest in that topic in the industry. But there are a very few repos and blogs about that. The repos that are out there doesn't cover many things in detail. This will help a lot of people and get good attention.

Hey!! Mr, can you make video on how to Deploy object detection model online, plz😇.. And love your content as always🤗🤗

Bro some girl has copied ur code and she is getting more fame than u.. I also came on ur video after seeing her video..

Amazing Nicholas, love all your videos. Had similar project in Matlab, but loved your python-tensorflow version the best...Amazing keep up the amazing video and content work!!!!

nice thank you for the tutorial

Hey! Thank you for this amazing tutorial, I'm trying to create a video conferencing project and want to include this as a feature, can you please guide me on how I can do that?

Hello Nicholas, I hope you’re doing well❤️ I would like to thank you for your video that helped us alot💘. Now I’m actually in the training step, I tried to train the model in my laptop “which doesn’t have a GPU “ so it took 10 hours to train 20 photos. Is there a way or a device that I can buy that will make the training process faster in my laptop?

How did you manage your time, both for you full-time tech job ( that might be overtime ), youtube project, and your life activities. Love your channel ❤

Thank you so much!

VIT wale full credit lekr marketing kr rhe. koi to roko unhe 😂

Hey there! I have issue with this "2. Create TF records" step. I am trying to run this and it is processing from a very long time without any output. Is there something I need to do ?

Thank you so much for this amazing video. I have one question. Can we add voice 'Hello' along with text 'Hello' in real-time detection? If yes please guide me. Thank you again.

Super useful!

[Errno 2] No such file or directory: 'Tensorflow/workspace/annotations\\label_map.pbtxt' I'm getting this error, can someone help me? (doing this on Collab, MacOS 10.15.3)

Please, solve me, SyntaxError: invalid syntax how to ?

hola gracias por el tutorial !

Hey, I'm so thankful for all the help you have given me. Thankyou so much. I am done with step 6 with the training part. When I ran the 7th step, I got this error: --------------------------------------------------------------------------- NameError Traceback (most recent call last) in 1 # Load pipeline config and build a detection model ----> 2 configs = config_util.get_configs_from_pipeline_file(CONFIG_PATH) 3 detection_model = model_builder.build(model_config=configs['model'], is_training=False) 4 5 # Restore checkpoint NameError: name 'config_util' is not defined What do I do?? Please let me know. I need help.

Sir, when i capture the images they are not showing in the collectedimage folder. Please help

Hi Nicholas! Thank you so much! I got this, please help: error: OpenCV(4.4.0) /tmp/pip-req-build-dglzv4yn/opencv/modules/imgcodecs/src/loadsave.cpp:738: error: (-215:Assertion failed) !_img.empty() in function 'imwrite'

Absolute YES!

only after i started working on this that i realise how much effort it for this 30min video

ValueError : 'images' must have either 3 or 4 dimensions Why :/

Hi nicholas. Thank you for the amazing tutorial!! Im facing a tiny problem and I need help. When I run the cell 'config = config_util.get_configs_from_pipeline_file(CONFIG_PATH)' I get the following error: 'module tensorflow has no attribute gfile' I tried replacing every tf.gfile in the config_util with tf.io.gfile and it still doesn't work. How can I fix this please?

you are a life saver, your video has put an end to my search on how to train my model on terminal instead of notebook. But my question is how can i plot the lost graph if i train my model in terminal

Woo awosm handy i am learning deep learning todY w8sh me luck bro i will try

Thanks for the tutorials. Really helps and they are amazing! Wanted to ask you what when you want to run it without having to use the commands at jupyter notebook. Like, just a normal app or computer program? Can you help me with that step?

Hi Nicho! Great tuto as usual. I wonder if it's possible to use videos instead of images for training because I noticed that many signs requires a sort of movement? Thanks in advance.

Hello Nicolas. I'm a fan of your channel, and I wanted to congratulate you for the video. Could you adapt this code to recognize water drops or do you think that another video of yours would be better adaptable? Thanks.

Nicholas, thank you, that's a 100% subscribe! May i ask you a little question, please? I know that optimal number of num_steps parameter depends on the task and on the model. Is there a way to implement early stopping in TFOD to get rid of this hyperparameter? Thank you so much, mate! Hello from Italy!

Hi - it sounds really exciting, I wondering if the model needs to be retrained to detect movement with a new person. I missed what type of information the model is using to make the detection so I am not so sure how much generalizable it would to different people with different hand/upper limb size, skin color or objects wore on the hand (ring, wristband, watch...).

Awesome mate

Incredible man. Loved it. Btw, what version of OpenCV did you use?

thanks so much sir

Thank you !

Hi there, were are making an android application for a graduation project and I'm wondering how to link this API to our code in android studio? in the end, I'm really glad that I found your video! thanks a lot ❤️!

Hello Nicholas. I have all the steps in your other video and trained the model. Our images have taken and labeled then moved to train and test folders inside the images folder. The model can recognize some photos but when use the webcam the model unable to detect and recognize any of the signs. Could you please gives some recommendations? We have 500 images for each of our 7 signs but we use other sign language not yours.

thank you for this great video, I have questions how I can use my existing dataset?