Skip to content

VisualAI is an mini artificial intelligence project written in Java and Angular that's made for training on custom image data and then object recognition within images.

Notifications You must be signed in to change notification settings

szivkovicx/visual-ai

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 

Repository files navigation

Java Angular Spring TailwindCSS TypeScript

image

VisualAI

VisualAI is an mini artificial intelligence project written in Java and Angular that's made for training on custom image data and then object recognition within images. It's flexible and easy to use.

image

Under the hood

VisualAI consists of 3 projects:

  • Model trainer ( /src/model directory ) - Mini neural network written in Java controlled completely using a single conf.json file ( see /src/model/conf.example.json for reference )
  • Model API ( /src/api directory ) - Mini API written using Spring Boot consisting of 2 routes, one ( GET @ /list ) written for listing all requests and another one ( POST @ /predict/{model_id} ) written for fetching probabilities estimated by neural network for each specific label
  • Model Web ( /src/web directory ) - Mini web application written in Angular made for listing these models and general usage easier

Neural network in VisualAI is as follows:

image

And is controlled by parameters provided in models conf.json file which describes basic meta of the model, training parameters, labels and features.

Example

Example below shows an cat/dog classifing model called animal-recognizer. To start, we are first going to train the model using 5 images of dogs and 5 images of cats which we are going to label cat and dog. We will be training this for 10 epochs with default seed of 1, normalization resolution of 500x500 and learning rate a bit over the usual and average one, 0.005. Configuration file ( conf.json ) will look like this:

{
    "name": "animal-recognizer",
    "version": "v0.1",
    "training_parameters": {
        "resolution": [
            500,
            500
        ],
        "learning_rate": 0.005,
        "seed": 1,
        "epochs": 10
    },
    "labels": [
        {
            "id": "dog",
            "data": [
                "../train_data/dog.1.jpg",
                "../train_data/dog.2.jpg",
                "../train_data/dog.3.jpg",
                "../train_data/dog.4.jpg",
                "../train_data/dog.5.jpg"
            ]
        },
        {
            "id": "cat",
            "data": [
                "../train_data/cat.1.jpg",
                "../train_data/cat.2.jpg",
                "../train_data/cat.3.jpg",
                "../train_data/cat.4.jpg",
                "../train_data/cat.5.jpg"
            ]
        }
    ]
}

Once we've done that, we can proceed by running train.cmd in the terminal. After some time ( depending on machine ), we should see an simple output like this:

image

Then, we are free to run our API and web application. We are going to run API by going to /src/api directory and running mvnw spring-boot:run. Then, we can go to /src/web and run npm start. We should see the following:

Screen-Recording-2024-09-09T08_55_27 362Z

About

VisualAI is an mini artificial intelligence project written in Java and Angular that's made for training on custom image data and then object recognition within images.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published