Great video, I am going to try this today. My first goal is to make a logic that can recognise if a image is of a cloud in the sky, or anything else. Thanks for your great video
Works like a charm, thank you for sharing! :D Also creating and using my own model worked well based on your sample code. I wonder one thing: why is the input image resized to 240x240 to "match the model"? And how do I unfluence this when creating the model? I did not find any option in Custom Vision to influence the dimensions. Also my input images to create the model all had 640x480, but the resulting model only seemed to work with the 240x240 resizing.
Thank you so much Sascha, been looking for a JavaScript ML CNN solution like this for a while. Please if you can create more JavaScript type tensor flow solutions like NLP or RNN. Thank you for this effort.
Welcome aboard! Since you are new, I would like to let you know if you have anything I should make a video about, just use the Hashtag #saschadittmann with your comment and I'll see what I can do. If you haven't already, check out my channel page to see all topics I've covered so far ua-cam.com/channels/bX8EdsCb-jFDfr05hzMjjw.html I look forward to seeing you around! 👊
I've been wanting to do something with image recognition, but more along the lines of "recognize faces from a specific set in this image" (in my use-case, not real faces but game-character avatars). Would Tensorflow be a good solution to that end? Could you maybe share some good entry level videos or links that might be relevant?
I've ran your github project on web browser(chrome, internet explorer, microsoft edge). It worked well until push 'predict' button. But when i push 'predict' button, output result didn't show up. Maybe it is concerned with server problem?
Did you download the code as zip file? If yes, this is causing the problem. I used Git LFS to store the model file. But you can download it separately here github.com/SaschaDittmann/tfjs-customvision/raw/master/static/model/weights.bin
@@SaschaDittmann now i am trying made image classification api. But have some troubles, input image must be base64 encoded image. When i try convert to 224 size, i have errors. Maybe u know what to do?
Hi, I tried to build on your existing code, and I got an error. Any help please? (all the details are on Stack overflow: stackoverflow.com/questions/61381112/uncaught-error-based-on-the-provided-shape-1024-3-the-tensor-should-have-30)
Hi Mack Cz, you will get such error messages if the bin file is corrupt. This can happen, for example, if you downloaded a repo from GitHub as a ZIP file, which uses Git LFS instead. I'm using Git Large File Support for uploading the model binaries. I saw this error before, when people downloaded my Git Repo as ZIP file. To avoid that you need to clone it or download the bin file separately. In your case I assume that your website hoster does not allow the file extension .bin and you might get an error message as content of the file back. Also the download times of the json and bin are a bit surprising. Both are almost identical, although the bin file should be much bigger and therefore requires a longer download.
And if I need an image generator? I already have the trained generator model,the output is a tensor of 128x128x3, how can I convert the output in a HTML image? Sorry for my English, I'm a Spanish native😂
If you need to more control over the training process, you can use either the regular TF framework and convert the model or train it in TFJS like I did here ua-cam.com/video/15yowpaXPM0/v-deo.html
Thank you so much for the video. Good job! Do you know if there is a way to convert RGBA images to grayscale images with TensorFlow.js? I am currently struggling with my own project where I need to get the image data from an html canvas and then convert it to grayscale before making the prediction. #saschadittmann Viele Grüße!
Hi Mahajan! What exactly do you want to do? Do you want to upload the pictures, for example via SDK? If yes, you can use the Custom Vision SDKs: docs.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/quickstarts/image-classification?pivots=programming-language-csharp
I used the node server just for debugging and running the project locally. But in the end, it's just a static webpage. You can use this approach in any JavaScript project.
Ok I saved my entire model with h5 format and downloaded json and 1 to 3 shaded file and so on. I didn't downloaded weight.bin file as I downloaded the entire model. Will my model be loaded using your code?
TensorFlowJS has its own model format. You have to convert your current model first to be able to use it with this framework. After that, you need to make sure, that you transform your incoming data (images) to fit your model input layer (like I did in my code) and set the appropriate labels. If you did these steps, my code should work.
Great video, I am going to try this today. My first goal is to make a logic that can recognise if a image is of a cloud in the sky, or anything else. Thanks for your great video
Sounds great! Let me know if you were able to create such a model.
Works like a charm, thank you for sharing! :D Also creating and using my own model worked well based on your sample code. I wonder one thing: why is the input image resized to 240x240 to "match the model"? And how do I unfluence this when creating the model? I did not find any option in Custom Vision to influence the dimensions. Also my input images to create the model all had 640x480, but the resulting model only seemed to work with the 240x240 resizing.
Hi Frank! As far as I know, you're not able to change the input size at this point in time. But maybe this will be possible in the future.
Thank you so much Sascha, been looking for a JavaScript ML CNN solution like this for a while. Please if you can create more JavaScript type tensor flow solutions like NLP or RNN. Thank you for this effort.
THANKS SIR, YOU SAVED MY LIFE
You're welcome! I really appreciate what you think!
What else do you think I could cover?
you mentioned alot about the tuberculosis detector but i cant seem to find that video
Thank you very much. I just subscribed
Welcome aboard! Since you are new, I would like to let you know if you have anything I should make a video about, just use the Hashtag #saschadittmann with your comment and I'll see what I can do.
If you haven't already, check out my channel page to see all topics I've covered so far ua-cam.com/channels/bX8EdsCb-jFDfr05hzMjjw.html
I look forward to seeing you around! 👊
I've been wanting to do something with image recognition, but more along the lines of "recognize faces from a specific set in this image" (in my use-case, not real faces but game-character avatars). Would Tensorflow be a good solution to that end? Could you maybe share some good entry level videos or links that might be relevant?
Hi! I planned to create a video on object detection as well. This might solve your issue.
I just want to show the results, what should I do?
Help me
I've ran your github project on web browser(chrome, internet explorer, microsoft edge). It worked well until push 'predict' button. But when i push 'predict' button, output result didn't show up. Maybe it is concerned with server problem?
Did you download the code as zip file? If yes, this is causing the problem.
I used Git LFS to store the model file.
But you can download it separately here github.com/SaschaDittmann/tfjs-customvision/raw/master/static/model/weights.bin
Good job sacha!
Thanks! I really appreciate what you think!
What else do you think I could cover?
@@SaschaDittmann Thank you, maybe how to train a CNN model locally with node.js
Thanks for sharing your thoughts!
I did already a video on that topic.
Please have a look: ua-cam.com/video/15yowpaXPM0/v-deo.html
@@SaschaDittmann Thank you Sascha I will check that
You‘re welcome! If you have any further ideas, don‘t hesitate to ping me.
Love u, thank u very much!
Thanks for letting me know. I really appreciate it!
Are there any other topics that interest you?
@@SaschaDittmann now i am trying made image classification api. But have some troubles, input image must be base64 encoded image. When i try convert to 224 size, i have errors. Maybe u know what to do?
can you give the link to where you made the TB detector
Hello, do you have a tutorial on how to implement your image classification tutorial to Node-Red?
I‘m sorry but no. But I‘m happy to add that to my list of ideas
como se cargan las imagenes o el modelo alguien me podria ayudar porfavor....
Wonderful video
It's possible to create a model tfjs without azure custom vision?
thank you
Yes, sure... Please checkout this video ua-cam.com/video/15yowpaXPM0/v-deo.html
Hi, I tried to build on your existing code, and I got an error. Any help please? (all the details are on Stack overflow: stackoverflow.com/questions/61381112/uncaught-error-based-on-the-provided-shape-1024-3-the-tensor-should-have-30)
BTW thanks so much for the tutorial. Really appreciate it
Hi Mack Cz, you will get such error messages if the bin file is corrupt.
This can happen, for example, if you downloaded a repo from GitHub as a ZIP file, which uses Git LFS instead.
I'm using Git Large File Support for uploading the model binaries.
I saw this error before, when people downloaded my Git Repo as ZIP file.
To avoid that you need to clone it or download the bin file separately.
In your case I assume that your website hoster does not allow the file extension .bin and you might get an error message as content of the file back.
Also the download times of the json and bin are a bit surprising. Both are almost identical, although the bin file should be much bigger and therefore requires a longer download.
i'm facing same issue can you explain how to do it, I didn't get it with stackoverflow
And if I need an image generator?
I already have the trained generator model,the output is a tensor of 128x128x3, how can I convert the output in a HTML image?
Sorry for my English, I'm a Spanish native😂
If you need to more control over the training process, you can use either the regular TF framework and convert the model or train it in TFJS like I did here ua-cam.com/video/15yowpaXPM0/v-deo.html
Can tensorflow js output an image with bounding box
Are you talking about object detection and drawing a rectangle around the objects?
@@SaschaDittmann yeah exactly
Yes, that's possible and I'm going to create a demo and video for this scenario too.
That will be very helpful thanks in advance
Thank you so much for the video. Good job! Do you know if there is a way to convert RGBA images to grayscale images with TensorFlow.js? I am currently struggling with my own project where I need to get the image data from an html canvas and then convert it to grayscale before making the prediction. #saschadittmann Viele Grüße!
Azure Microsoft need payment method.
Best. Is there is another way to create a dataset?
Hi Mahajan!
What exactly do you want to do?
Do you want to upload the pictures, for example via SDK?
If yes, you can use the Custom Vision SDKs:
docs.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/quickstarts/image-classification?pivots=programming-language-csharp
Is it possible to build tensorflow project without the node server?
It is possible. You can run with React or localhost.
I used the node server just for debugging and running the project locally.
But in the end, it's just a static webpage.
You can use this approach in any JavaScript project.
How can I configure VS code to use it with tenserflow.js ?
What exactly are you looking for? JavaScript works out of the box, but you can definitely add some extensions to make it easier to work with.
I was getting the error
$,reference not found
The $ is from jquery and not TFJS.
Could you please check in the browser dev tools, if jquery is loaded successfully.
Ok I saved my entire model with h5 format and downloaded json and 1 to 3 shaded file and so on. I didn't downloaded weight.bin file as I downloaded the entire model.
Will my model be loaded using your code?
TensorFlowJS has its own model format. You have to convert your current model first to be able to use it with this framework.
After that, you need to make sure, that you transform your incoming data (images) to fit your model input layer (like I did in my code) and set the appropriate labels.
If you did these steps, my code should work.