i want create model in js that could recognize some made up icons from an image and i am confused with all the data, what is the best way to do it ? using pre model ? which one ? thank you very much 
Are you trying to do image recognition or object recognition on those icons?
Image recognition being knowing if an icon exists in an image vs object recognition that puts a bounding box around each one and tries to guess their position in the image too?
I don’t care where are they, just how many they are and which colors they have, thank you for your answer 
Then you need to use Object Recognition and I suggest you follow the following tutorial:
or this one:
TensorFlow has a bunch of great options for this, but I’d recommend looking into TensorFlow.js since you’re working with JavaScript.
For pre-trained models, you might want to check out the MobileNet models. They’re lightweight and work well for image classification tasks. TensorFlow.js even has a handy converter to convert TensorFlow models to a format that works in the browser.
If you want to choose a pre-trained model, MobileNet in my opinion is a good starting point. If your icons are a bit complex, you might want to explore more advanced models like ResNet.
Do let me know if it has helpful @yigalipsey1996
thank you very much
Mobilenet alone will not solve their problem as they need to do object detection not image recognition as they need to count how many objects are found. Once object is found they would need to then sample colour to find colour etc.
can i train model to recognize the icons directly from js ? i struggle top
do it via python , if you have good tutorial for this i will be happy , thank you
when i follow the first tutorial in the training command i get this error " AttributeError: module ‘tensorflow.python.ops.control_flow_ops’ has no attribute ‘case’"