https://github.com/goamegah/fastapi-self-supervised-learning-images

API in order to serve S3Ima models
https://github.com/goamegah/fastapi-self-supervised-learning-images

backend fastapi image-processing python pytorch self-supervised-learning

Last synced: about 1 month ago
JSON representation

API in order to serve S3Ima models

• Host: GitHub
• URL: https://github.com/goamegah/fastapi-self-supervised-learning-images
• Owner: goamegah
• License: apache-2.0
• Created: 2023-12-04T17:17:06.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-12-10T16:47:27.000Z (about 1 year ago)
• Last Synced: 2024-10-17T12:25:42.608Z (3 months ago)
• Topics: backend, fastapi, image-processing, python, pytorch, self-supervised-learning
• Language: Python
• Homepage:
• Size: 3.32 MB
• Stars: 3
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# SsIma API

SsIma API aim to implement API in order to serve all model implement in [s3ima](https://github.com/goamegah/s3ima).

After, I containerize api using docker.

## Table des Matières

1. [Introduction](#introduction)

2. [API documentation interface](#api-documentation-interface)

2. [Project workflows](#project-workflows)

3. [Run model](#run-api)

---

## Introduction

Project concerns of implementing simple deep learning architecture

for classifying ```MNIST``` image data with two constraints

- Train model with only 100 labels samples:

- Implement API to serve model functionality

- Implement Web App (as client) to display model performance.

**API** will be used to serve model by exposing two functionalities:

- Receiving request that will contains image from a **Frontend app** 

which (assuming) to be digit image and ```model name```.

- Response by posting **inference result** and ***model evaluation metrics*** as graph.

Project workflow below, will help user understanding where APi will take part

and its roles. (Comments use French language)

---

**Attention:** _This part of project concerns only API implementation and deployment as service

using Docker_. 

---

**Requirements for Running whole process (Front-API-Model):**

- For having a look on model's implementation, you will need to refere to [s3ima](https://github.com/goamegah/s3ima)

project. Obviously, you don't need to have s3ima code for testing API due to prior availability of artefact inside 

**artefact/** folder.

- You will probably need to have Dash web app implementation on local. So you can get code 

from [s3ima-app](https://github.com/goamegah/ssima-app)

## Api documentation interface:

Here is how API interface will look like.

![Alt Text](documentation/fast_api_docs.png)

## Project workflows:

![Alt Text](documentation/local_arch.png)

Dash web app (ssima-app) depend on API (s3ima-api) results from model (s3ima). 

To simplify process all artefact is prior available through **artefact/** folder. 

Api server will be waiting for request (eg. image) send by front (Dash App) and 

will get and process image, feed model, get result from it and finally sent results 

to the front (dash web app) .

![Alt Text](documentation/dash_screen.png)

## Run API

Please before going on, make sure having installed ```uvicorn```, ```python-multipart``` and ```torch```.

If not, you can require librairy by run ```pip install -r requirements.txt```

In the root of project can launch model API by using following command

```shell

$  uvicorn api:app --reload

```

After this you can open your navigator and tape this link: ```http://127.0.0.1:8000/``` 

## Containerize app with Docker

for containerize API app, **dockerfile** is available for building and deploy image

```shell

$ docker build -t image_name:version .

$ docker run -p 8000:8000 image_name:version

```