def _add_bbox_to_image()

in src/nova_act/impl/run_info_compiler.py [0:0]

• 16 lines of code
• 5 McCabe index (conditional complexity)

def _add_bbox_to_image(image: str, response: str) -> str:
    if not image:
        return image

    # Find the first bbox in the response. Right now there can ever only be on bbox. The agent will only take one
    # action at a time and then observe before taking the next one.
    bbox_match = _BBOX_MATCHER.search(response)
    if not bbox_match:
        return image
    top, left, bottom, right = map(int, bbox_match.groups())

    # Strip the data prefix in the base64 image.
    image_match = _IMAGE_PREFIX_MATCHER.match(image)
    if image_match:
        image = image_match.group(1)

    # Add the rectangle to the image.
    pil_image = Image.open(io.BytesIO(base64.b64decode(image)))
    draw = ImageDraw.Draw(pil_image)
    draw.rectangle((left, top, right, bottom), outline="red", width=3)
    image_bytes_io = io.BytesIO()
    pil_image.save(image_bytes_io, format="JPEG")

    # Return the modified image with the data prefix.
    return "data:image/jpeg;base64," + base64.b64encode(image_bytes_io.getvalue()).decode("utf-8")