alexmudrak / image-ml-classify

Image ML Classify is a versatile web application for image processing and ResNet-18 model training. Store and sync datasets on Yandex Disk. Utilizes Docker for background model training, ensuring data availability and convenience.

Image ML Classify - Image Processing and ResNet-18 Model Training Tool

Image ML Classify is a versatile web application that serves as a powerful image processing and ResNet-18 model training tool. It provides the convenience of storing and synchronizing datasets in the Yandex Disk cloud storage.

The application utilizes Docker for containerization, and model training occurs in the background, freeing the main thread for other tasks. All datasets can be stored and synchronized with Yandex Disk, ensuring data availability and convenience for model training.

Endpoints

Below are the endpoints and sample CURL requests for using this tool:

Image Classification

Upload an image and obtain classification results using the pre-trained ResNet-18 model.

CURL Request:

curl -X POST \
-F "file=@your_image.jpg" \
http://your-api-url/api/v1/image

Model Training

Initiate the training of the ResNet-18 model with your custom datasets in the background.

CURL Request (GET - Get Current Training Status):

curl -X GET \
-H "X-Key: your_api_key" \
http://your-api-url/api/v1/train

CURL Request (POST - Start Training):

curl -X POST \
-H "X-Key: your_api_key" \
-H "Content-Type: application/json" \
-d '{"epoch": 10, "hard_run": false, "run_as": "all"}' \
http://your-api-url/api/v1/train

Getting credentials for Yandex.Disk

How-To?

Requirements:

• Create a new application (optional, you can use a third-party application; you'll need its ID and secret)
  • App ID
  • App Secret
• Authorize the application to access the disk
  • Get the authorization code
  • Get the access token

Create an Application (Optional)

Official instructions - https://yandex.ru/dev/disk/api/concepts/quickstart.html#quickstart__oauth

The main goal is to obtain the App ID and App Secret; you can use an existing application.

Obtaining an Authorization Code

Navigate to the following link - https://oauth.yandex.ru/authorize?response_type=code&client_id={APP_ID} You will be asked to confirm permissions for the application.

After redirection, you will be provided with a code http://.../?code=7122172 the authorization code is valid for 15 minutes

Obtaining a Token

After obtaining the authorization code, you can acquire the application's authorization token. To do this, make a POST request, providing the necessary data

curl -X POST \
-L "https://oauth.yandex.ru/token" \
-F "grant_type=authorization_code" \
-F "client_id={APP_ID}" \
-F "client_secret={APP_SECRET}" \
-F "code={AUTH_CODE}" \

image-ml-classify's Issues

Add implementation for new datamodel

Refactor code

• Create class Config
• Add auto-generated value in config
• Config for PATH classes,json
• Config for PATH status DB

Add documentation

• How to add yandex Disk token
• Complex methods

Add tests

Add yandex.disk client

Implement setter to epoch count

Implement the setup count number of an epoch of the model train by sending a request to the train endpoint.

Implement logger and add log

Replace all print to logger

Dockerize app

Implement and fix types

Refactor and add endpoints

• Get class
• Run train