Melanoma Binary Segmentation Lab

This FastAPI service allows clients to upload an image and perform neural network (NN) inference on it. The server resizes the image, processes it through a specified model, and returns the inference results as a PNG image.

Build documentations

sphinx-build -b html docs/source docs/build

And in docs/build you may find generated docs

Installation

Docker

Just build and run docker with bash file- it will build and run container
```
sh build_docker.sh
```

Manual

Create and activate a virtual environment:

python3 -m venv venv  
source venv/bin/activate

Install the required dependencies:
```
pip install -r requirements.txt  
```

Running the Server (for manual usage)

To start the FastAPI server, run:

uvicorn application:app --log-level info

The server will be available at http://127.0.0.1:8000.

API Endpoints

`/nn_inference`

Description

This endpoint accepts an image file and a neural network type, performs inference using the specified model, and returns the processed image.

Request

Method: POST
Content-Type: multipart/form-data

Parameters

Name	Type	Description
`file`	`UploadFile`	The BMP image file to be processed.
`nn_type`	`Form`	The type of neural network model to use for inference.

choose nn_type from ("segnet", "unet", "naive"), by default- segnet

Example cURL Request

curl -X POST "http://127.0.0.1:8000/nn_inference" \  
-F "file=@path/to/your/image.bmp" \  
-F "nn_type=model_type"
-o segmentation.png

Response

Content-Type: image/png
Returns the image with inference results.

Example of usage

Via bash file client_curl.bash which equal next:

curl -X 'POST' \
  'http://127.0.0.1:8000/nn_inference' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@IMD036.bmp' \
  -F "nn_type=unet" \
  -o segmentation.png

Via python script:

source venv/bin/activate && python client_python.py

Logging

The server logs the following information:

Shape of the processed image.
Type of the model used for inference.
Shape of the inference results.

Additional Notes

Ensure that the get_container, get_model_from_container, read_file, resize, imread, create_img, and logger functions/utilities are defined and properly implemented in your codebase.
The example assumes that the image processing libraries (e.g., skimage, torch, numpy) are installed and properly configured.

Contributing

Contributions are welcome! Please submit a pull request or open an issue to discuss any changes or improvements.

License

This project is licensed under the MIT License.

   
This `README.md` provides a comprehensive overview of the FastAPI endpoint, including installation instructions, API usage, and detailed explanations of the endpoint's functionality. Make sure to adjust paths, URLs, and other details according to your actual project setup.

Project tree

.
├── .env
├── .gitignore
├── Dockerfile
├── IMD036.bmp
├── README.md
├── application.py
├── artifacts
│   └── models
│       ├── NaiveCNN
│       │   └── NaiveCNN_dice_250e.pt
│       ├── SegNet
│       │   └── SegNet_dice_250e.pt
│       ├── SegNet_1
│       │   └── SegNet_dice_250e.pt
│       └── UNet
│           └── UNet_dice_250e.pt
├── build_docker.sh
├── clients
│   ├── client_curl.bash
│   └── client_python.py
├── docs
│   ├── Makefile
│   ├── make.bat
│   └── source
│       ├── _static
│       ├── _templates
│       ├── application.rst
│       ├── conf.py
│       ├── index.rst
│       ├── nn_models.rst
│       ├── provider.rst
│       ├── train_losses.rst
│       ├── trainer.rst
│       └── utils.rst
├── requirements.txt
├── src
│   ├── __init__.py
│   ├── data
│   │   ├── __init__.py
│   │   └── provider
│   │       ├── gdrive_weights.py
│   │       ├── iph2_provider.py
│   │       ├── ph2_provider.py
│   │       └── provider_dependency_injection.py
│   ├── endpoints
│   │   ├── __init__.py
│   │   └── nn_inference.py
│   ├── metrics
│   │   ├── __init__.py
│   │   ├── losses
│   │   │   ├── __init__.py
│   │   │   ├── bce_loss.py
│   │   │   ├── dice_loss.py
│   │   │   ├── focal_loss.py
│   │   │   └── iloss.py
│   │   ├── metrics_dependency_injection.py
│   │   └── quality
│   │       ├── __init__.py
│   │       └── iou.py
│   ├── nn_models
│   │   ├── __init__.py
│   │   ├── naive_cnn.py
│   │   ├── nn_dependency_injection.py
│   │   ├── segnet.py
│   │   └── unet.py
│   ├── trainer
│   │   ├── __init__.py
│   │   ├── trainer.py
│   │   └── trainer_dependency_injection.py
│   └── utils
│       ├── get_model.py
│       └── read_files.py
└── train.py

IshchenkoRoman/CV