Custom object detection on webcam with Tensorflow.js

Hello, I’m new to using tensorflow.js. I want to do custom object detection using Tensorflow.js. I tried to perform the work done by Mr. Hugo Zanini on the article “Custom Real-Time Object Detection in the Browser Using TensorFlow.js” on the Kangaroo Dataset, but I am getting some errors. Is there an article or video to help us do custom object detection using Tensorflow.js?

Yes, there are problems with version mismatch in the kangaroo detector. There is another blog by Hugo Zanini where he is detecting SKU by training a custom model SKU Detection, however when I tested this blog’s sandbox link, it’s only working with the current model trained not with other models and it gave me error message “info is undefined” and "tensor is disposed’. however, you are welcome to give this one a try.

1 Like

I will definitely try this work out, thank you.

Hi @Semra_Yildirim, welcome to the forum.

Feel free to have a first look here:
Depending on your needs, it should have the same achievements like in the article (just without coding).

If you would like to deep dive a bit more into tensorflow.js, this might be a fantastic start (you will learn how to program your own custom object detection from scratch in JS).

1 Like

@hugozanini FYI possibly some errors on your blog example - let me know if we need to update anything

1 Like

A similar problem has been discussed here: TFJS model not working problem - #5 by hugozanini. I got a model by completing the training, but when I run the model I converted in the browser, I get errors. I can open it in the browser with the application in the SKU example, but it is perceived as the SKU model. I tried the “saved_model_cli show --dir {mobilenet_save_path} --tag_set serve” code in the link, but I got the following output:
The given SavedModel MetaGraphDef contains SignatureDefs with the following keys:
SignatureDef key: “__saved_model_init_op”
SignatureDef key: “serving_default”

Thank you very much for your interest.

1 Like

@Semra_Yildirim @Mayukhdeep @Dennis @Jason

Hi, everyone

Thanks for using the post and tagging me. I have just run the sandbox link and it’s working properly.

@Semra_Yildirim Are you transfer learning a MobileNet V2 model? Tensorflow.js has a list of supported layers and from what you sent, it looks like an error related to to the use of an unsupported op.


Thanks Hugo for checking this one out.

1 Like