whitphx / streamlit-fesion

streamlit-fesion

Streamlit component for frontend computer vision processing with Wasm/Pyodide.

Sister project: streamlit-webrtc

---

You write an image filter function in Python that receives an image frame, transforms it, and returns a processed image frame. streamlit-fesion exports it to the frontend environment and executes it with Pyodide to process the WebCam video stream.

Look at the code below. It is a grayscale filter example.

from streamlit_fesion import streamlit_fesion


def image_filter(input_image):
    import skimage

    grayscale = skimage.color.rgb2gray(input_image)
    return skimage.color.gray2rgb(grayscale)


streamlit_fesion(image_filter, [], key="fesion")

image_filter() is the filter function. streamlit-fesion will call it with an input image frame of type np.ndarray with 3 channels (RGB-ordered), and the filter function returns a processed image frame with the same type and shape.

Note that the image_filter() will be sent to the frontend environment and executed there, but any other parts of the code will not. Therefore, the packages used in the filter function must be imported inside it, like import skimage in the example above.

streamlit-fesion automatically detects the imported packages and installs them to the frontend environment at the initialization time¹. However, if necessary, you can explicitly pass the requirements list to the second argument of streamlit_fesion(), where an empty list [] is passed in the example above.

The signature of streamlit_fesion() follows.

streamlit_fesion(
    filter_func: Callable[[np.ndarray], np.ndarray],
    dep_packages: Optional[List[str]] = None,
    key: Optional[str] = None,
)

Footnotes

1. pyodide.loadPackagesFromImports is used for it. ↩