arda-num/Training-Server

Training-Server

This is a AI Model training and store server made by FAST API.

Example

http get request example in browser to train and store:

http://127.0.0.1:8000/progress?train_type=classification&bucket_name=client0

Important: bucket_name represents user specific bucket name on the cloud

File Organization in Local

root (dir) 
    main.py (file)
    train_data (dir)
        client0 (dir)
            coco (dir) 
                annotations.json
            images (dir)
                images...
        client1 (dir)
            coco (dir)
                annotations.json
            images (dir)
                images...
    weights (dir)
        client0 (dir)
            model.h5 (file)
        client1 (dir)
            model.h5 (file)

This represents a general idea of how documentation is done. When a new user requests for a training new directory added for that user. On the other hand, when an existing user requests, his/her directory is deleted and generated from scratch. Also notice that models are just appear before the upload operation. After the upload is done, models are deleted from local.

Sample annotations.json file (COCO) for Classification:

{
    "info" : {},
    "licences" : [],
    "categories" : [
        {
            "id" : 1,
            "name" : "cat"
        }
    ],
    "images" : [
        {
            "id" : 0,
            "licence" : 1,
            "file_name" : "00000001.jpg",
            "height" : 324,
            "width" : 765
        }
    ],
    "annotations" : [
        {
            "id" : 3,
            "image_id" : 0,
            "category_id" : 1,
        }
    ]
}

• annotations.json file must be like the above sample; otherwise, server will give an error.

File and Bucket Organization in S3

root
    training-server-client0 (Bucket)
        coco (dir)
            annotations.json
        images (dir)
            images...
    training-server-client1 (Bucket)
        coco (dir)  
            annotations.json
        images (dir)
            images...
                .
                .
                .
    training-server-model-client0 (Bucket)
        model.h5 (file)
    training-server-model-client1 (Bucket)
        model.h5 (file)
            .
            .
            .