Create an Image Recognition Example in JS TensorFlow -- 2

I want a working prototype of a JS TensorFlow project that I can use to train it to recognize (binary) images (not categorize images).

I can upload my own test examples and labels to train a network and then use it to validate un-seen images. It should basically be a probability of a valid image.

I will provide images and labels which basically say "this image is ok", "this image is not ok" to make the prototype work, but eventually I need to use this with new training data to train new networks. Always of the same type "Is this image ok or not?"

The Images I will provide need to be normalized. Once the product is delivered, new training sets will use normalized pictures as according to the specifications of the network.

Usage scenario A - Training

1) Fire up the project

2) Load a set of labeled examples used to train the network

3) Be able to save the trained network for later-processing

Usage Scenario B - Predicting

1) Fire up the project

2) Somehow choose a trained network (may have multiple)

3) Predict a single unlabeled example

Types of networks to train (all binary):

- "is this person dressed correctly?"

- "is this product correctly installed?" (i.e. a painting is straight or not)

- "is this glass empty or full?"

etc. I have training sets for the scenarios I want to test.

Umetna inteligenca JavaScript Machine Learning (ML) Tensorflow

ID projekta: #18085094

Več o projektu