svision_client.py

import os
os.environ["CUDA_VISIBLE_DEVICES"] = "1"

from gradio_client import Client, handle_file
from typing import Any, Dict, List, Optional, Tuple, Union
import json

# Connect to the remote Space
svision_client = Client("VeuReu/svision")


def extract_scenes(video_path: str, threshold: float = 30.0, offset_frames: int = 5, crop_ratio: float = 0.1):
    """
    Call the /scenes_extraction endpoint of the remote Space VeuReu/svision.

    Parameters
    ----------
    video_path : str
        Path to the input video file.
    threshold : float, optional
        Scene change detection threshold; higher values make detection less sensitive.
    offset_frames : int, optional
        Number of frames to include before and after a detected scene boundary.
    crop_ratio : float, optional
        Ratio for cropping borders before performing scene detection.

    Returns
    -------
    Any
        Response returned by the remote /scenes_extraction endpoint.
    """
    result = svision_client.predict(
        video_file={"video": handle_file(video_path)},
        threshold=threshold,
        offset_frames=offset_frames,
        crop_ratio=crop_ratio,
        api_name="/scenes_extraction"
    )
    return result  


def keyframes_every_second_extraction(video_path: str):
    """
    Call the /keyframes_every_second_extraction endpoint of the remote Space VeuReu/svision.

    Parameters
    ----------
    video_path : str
        Path to the input video file.

    Returns
    -------
    Any
        Response returned by the remote /keyframes_every_second_extraction endpoint.
    """
    result = svision_client.predict(
        video_path={"video": handle_file(video_path)},
        api_name="/keyframes_every_second_extraction"
    )
    return result


def add_ocr_and_faces(imagen_path: str, informacion_image: Dict[str, Any], face_col: List[Dict[str, Any]]) -> Dict[str, Any]:
    """
    Call the /add_ocr_and_faces endpoint of the remote Space VeuReu/svision.

    This function sends an image together with metadata and face collection data
    to perform OCR, face detection, and annotation enhancement.

    Parameters
    ----------
    imagen_path : str
        Path to the input image file.
    informacion_image : Dict[str, Any]
        Dictionary containing image-related metadata.
    face_col : List[Dict[str, Any]]
        List of dictionaries representing detected faces or face metadata.

    Returns
    -------
    Dict[str, Any]
        Processed output containing OCR results, face detection data, and annotations.
    """
    informacion_image_str = json.dumps(informacion_image)
    face_col_str = json.dumps(face_col)
    result = svision_client.predict(
            image=handle_file(imagen_path),
            informacion_image=informacion_image_str,
            face_col=face_col_str,
            api_name="/add_ocr_and_faces"
    )
    return result


def extract_descripcion_escena(imagen_path: str) -> str:
    """
    Call the /describe_images endpoint of the remote Space VeuReu/svision.

    This function sends an image to receive a textual description of its visual content.

    Parameters
    ----------
    imagen_path : str
        Path to the input image file.

    Returns
    -------
    str
        Description generated for the given image.
    """
    print("Calling svision to describe the scene...")
    result = svision_client.predict(
        images=[{"image": handle_file(imagen_path)}],
        api_name="/describe_images"
    )
    return result