Bounding box detection

Experimental

This feature is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of the Service Specific Terms. Pre-GA features are available "as is" and might have limited support. For more information, see the launch stage descriptions.

To see an example of bounding box detection, run the "Spatial understanding with Gemini 2.0" Jupyter notebook in one of the following environments:

Open in Colab | Open in Colab Enterprise | Open in Vertex AI Workbench user-managed notebooks | View on GitHub

In this experimental launch, we are providing developers with a powerful tool for object detection and localization within images and video. By accurately identifying and delineating objects with bounding boxes, developers can unlock a wide range of applications and enhance the intelligence of their projects.

Key Benefits:

• Simple: Integrate object detection capabilities into your applications with ease, regardless of your computer vision expertise.
• Customizable: Produce bounding boxes based on custom instructions (e.g. "I want to see bounding boxes of all the green objects in this image"), without having to train a custom model.

Technical Details:

• Input: Your prompt and associated images or video frames.
• Output: Bounding boxes in the [y_min, x_min, y_max, x_max] format. The top left corner is the origin. The x and y axis go horizontally and vertically, respectively. Coordinate values are normalized to 0-1000 for every image.
• Visualization: AI Studio users will see bounding boxes plotted within the UI. Vertex AI users should visualize their bounding boxes through custom visualization code.

Gen AI SDK for Python

Install

pip install --upgrade google-genai

To learn more, see the SDK reference documentation.

Set environment variables to use the Gen AI SDK with Vertex AI:

# Replace the `GOOGLE_CLOUD_PROJECT` and `GOOGLE_CLOUD_LOCATION` values
# with appropriate values for your project.
export GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
export GOOGLE_CLOUD_LOCATION=global
export GOOGLE_GENAI_USE_VERTEXAI=True

import requests
from google import genai
from google.genai.types import (
    GenerateContentConfig,
    HarmBlockThreshold,
    HarmCategory,
    HttpOptions,
    Part,
    SafetySetting,
)
from PIL import Image, ImageColor, ImageDraw
from pydantic import BaseModel

# Helper class to represent a bounding box
class BoundingBox(BaseModel):
    """
    Represents a bounding box with its 2D coordinates and associated label.

    Attributes:
        box_2d (list[int]): A list of integers representing the 2D coordinates of the bounding box,
                            typically in the format [y_min, x_min, y_max, x_max].
        label (str): A string representing the label or class associated with the object within the bounding box.
    """

    box_2d: list[int]
    label: str

# Helper function to plot bounding boxes on an image
def plot_bounding_boxes(image_uri: str, bounding_boxes: list[BoundingBox]) -> None:
    """
    Plots bounding boxes on an image with labels, using PIL and normalized coordinates.

    Args:
        image_uri: The URI of the image file.
        bounding_boxes: A list of BoundingBox objects. Each box's coordinates are in
                        normalized [y_min, x_min, y_max, x_max] format.
    """
    with Image.open(requests.get(image_uri, stream=True, timeout=10).raw) as im:
        width, height = im.size
        draw = ImageDraw.Draw(im)

        colors = list(ImageColor.colormap.keys())

        for i, bbox in enumerate(bounding_boxes):
            # Scale normalized coordinates to image dimensions
            abs_y_min = int(bbox.box_2d[0] / 1000 * height)
            abs_x_min = int(bbox.box_2d[1] / 1000 * width)
            abs_y_max = int(bbox.box_2d[2] / 1000 * height)
            abs_x_max = int(bbox.box_2d[3] / 1000 * width)

            color = colors[i % len(colors)]

            # Draw the rectangle using the correct (x, y) pairs
            draw.rectangle(
                ((abs_x_min, abs_y_min), (abs_x_max, abs_y_max)),
                outline=color,
                width=4,
            )
            if bbox.label:
                # Position the text at the top-left corner of the box
                draw.text((abs_x_min + 8, abs_y_min + 6), bbox.label, fill=color)

        im.show()

client = genai.Client(http_options=HttpOptions(api_version="v1"))

config = GenerateContentConfig(
    system_instruction="""
    Return bounding boxes as an array with labels.
    Never return masks. Limit to 25 objects.
    If an object is present multiple times, give each object a unique label
    according to its distinct characteristics (colors, size, position, etc..).
    """,
    temperature=0.5,
    safety_settings=[
        SafetySetting(
            category=HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
            threshold=HarmBlockThreshold.BLOCK_ONLY_HIGH,
        ),
    ],
    response_mime_type="application/json",
    response_schema=list[BoundingBox],
)

image_uri = "https://storage.googleapis.com/generativeai-downloads/images/socks.jpg"

response = client.models.generate_content(
    model="gemini-2.5-flash",
    contents=[
        Part.from_uri(
            file_uri=image_uri,
            mime_type="image/jpeg",
        ),
        "Output the positions of the socks with a face. Label according to position in the image.",
    ],
    config=config,
)
print(response.text)
plot_bounding_boxes(image_uri, response.parsed)

# Example response:
# [
#     {"box_2d": [6, 246, 386, 526], "label": "top-left light blue sock with cat face"},
#     {"box_2d": [234, 649, 650, 863], "label": "top-right light blue sock with cat face"},
# ]