Is there a type of A.I api that can be trained against recognizing special pixel images that do not contain regular images or faces or objects but things I train it to recognize? I want to be able to call this new trained model with an api. The idea is to avoid creating this using tensorflow.
Thank you.
Aside from TensorFlow, maybe dig into and modify something like:
Something from those might be a starting point for you. Unless I am misinterpreting your desires.
I want a Deep Learning Online API that I can train with my own way without writing it myself. Think, IBM Watson for instance. The problem with Google Vision and Azure Vision and IBM Watson is they can only be trained agains real objects and not specific objects or “things” that are out of the ordinary that I want to specify. I want to avoid using tensorflow for now until I have a working prototype.
I assume you found no interest in any of the above as starting points then. It might be a hard find given those two criteria. But I am interested to know if something exists that fits your needs out of the box.
What I thought about is something similar to IBM Watson, were you simply hand it a couple of pictures that you classify and it will learn based on that. When it has been trained, you simply call the API to use it. It is based on Deep learing. The libraries you mentioned are not based on Deep Learning.
I know, but …
So I’m looking forward to hear what you end up using. 
I have contacted Both Google And Watson about this. It seems like their A.I, only recognize regular objects and people. You cannot train it while giving it your own training data. That has to be programmed using Tensor Flow or something similar.
It is actually logical because they use a lot of time and effort to train their model. People would simply call their API with an image and a result is returned. They do not offer a simple way of letting the customer train it´s data. You simply have to write it yourself.
Thats weird thats not what the Watson site states (or what you mentioned earlier either).
To my understanding all of the following seem to declare custom models or input can be deployed?
I’m really curious what your looking to build.
This is strange, Watson support told me Watson does not support learning new images that have special pixels and such. It can be train agains regular images. I might have to look more. Do you work with Deep learning?
No not really.
Do you have an example? I’m just confused why you feel none of the existing (some mentioned above) would not work.
It is similar to the MINST dataset but with RGB colors 
I did not continue and just started doing other work. 
:: Copyright KIRUPA 2024 //--