https://github.com/stevekirks/py-imageclassification
A web api that uses the CLIP model and a list of categories to classify images.
https://github.com/stevekirks/py-imageclassification
Last synced: about 2 months ago
JSON representation
A web api that uses the CLIP model and a list of categories to classify images.
- Host: GitHub
- URL: https://github.com/stevekirks/py-imageclassification
- Owner: stevekirks
- License: mit
- Created: 2025-01-17T22:38:39.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2025-01-27T02:41:56.000Z (4 months ago)
- Last Synced: 2025-02-11T18:45:43.236Z (3 months ago)
- Language: Python
- Size: 210 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Image Classification API
A web api that uses the [CLIP](https://github.com/openai/CLIP) model and a list of categories (see `app/image_service.py`) to classify images.
It runs in a docker container with a volume for the model. This has been tested to run in an Azure Web App for Containers (linux) and performance is decent (500kb images < 0.4sec) on P0v3 (195 ACU/vCPU, 1vCPU, 4GB memory, AUD$73).
## Usage
### Local development with Dev Containers
For local development, you can use Docker and DevContainers in VS Code (requires the [Dev Containers VSCode extension](https://marketplace.visualstudio.com/items?itemName=ms-vscode-remote.remote-containers)).
With Docker running, clone the repo, open vscode, and if doesn't already prompt you, run the `Dev Containers: Reopen in Container` command.
Alternatively, you can setup a python virtual environment and install the packages in `requirements.txt`.
### Download the CLIP model
Before running the tests or web api, download the model. Run the python script `save_model_locally.py` to download the CLIP model to the ./data folder.
### Tests
There's some tests using `pytest`. See `tests/test_model.py`.
### To run in Docker without an IDE
```powershell
docker build --tag imgdet:latest .
docker run --rm -p 8080:80 -v my_repo_location/data:/volume imgdet
```
### Jupyter Notebooks
There's a jupyter notebook, `tests/analyse.ipynb`, that can be used for testing the web api.
### Azure
For deployment to an Azure Web App for Containers, these az cli commands can help:
Build image and push to container registry:
```powershell
az acr build --registry myregistry --image myregistry.azurecr.io/imgdet:latest .
```
Get web app to use latest container. Hacky (to avoid thinking about versions).
```powershell
az webapp config container set --name mywebapp --resource-group myresourcegroup --container-image-name myregistry.azurecr.io/imgdet:0.0.999
az webapp config container set --name mywebapp --resource-group myresourcegroup --container-image-name myregistry.azurecr.io/imgdet:latest
```
This isn't all the setup required for an Azure deployment. You'll also need to create an Azure File Storage resource, upload the model, and setup a `/volume` path in the web app pointing to it. Update the `model_path` in `main.py` to match.