Roboflow Docs

AWS S3 Bucket

Upload Images from an AWS S3 bucket to Roboflow

For AWS S3, use Datasources and Bucket Mirror unless you explicitly need a one-time or ad hoc import. Datasources keep images and metadata in sync without scripts, presigned URLs, or local downloads.

This page covers fallback workflows for one-time or ad hoc uploads only:

  • Signed URLs: Upload images directly from S3 to Roboflow using presigned URLs, without downloading them locally first.
  • CLI Locally: Download images from S3 to your local machine, then upload them to Roboflow. Use this when you need to preprocess or inspect images before uploading.

If you are setting up a reusable connection between Roboflow and S3, stop here and use Datasources plus AWS S3 Credentials.

AWS CLI Setup

Before using the script, ensure AWS CLI has access to the target bucket.

For local one-time workflows, use a non-root AWS identity with the minimum bucket permissions you need. If your goal is ongoing S3 ingestion into Roboflow, do not reuse this local setup. Create a Datasource credential instead.

Configuring the AWS CLI

  1. Once you've installed the AWS CLI, open a terminal or command prompt.

  2. Run the following command:

    aws configure

  3. You'll be prompted to enter your AWS credentials:

    AWS Access Key ID [None]: YOUR_ACCESS_KEY
AWS Secret Access Key [None]: YOUR_SECRET_ACCESS_KEY
Default region name [None]: YOUR_PREFERRED_REGION (e.g., us-west-1)
Default output format [None]: json

Option 1: Upload Via Signed URL

Use this only for one-time or scripted imports where you do not want to create a Datasource.

You can generate signed URLs for your images in the S3 bucket by using boto3 in Python.

def generate_presigned_url(bucket_name: str, object_name: str, region: str = 'us-east-2') -> str:
    """Generate a presigned URL for an S3 object."""
    s3 = boto3.client('s3', region_name=region, config=Config(signature_version='s3v4'))
    url = s3.generate_presigned_url('get_object',
                                    Params={'Bucket': bucket_name, 'Key': object_name},
                                    ExpiresIn=3600)
    return url

In the code snippet above, you need the name of your S3 bucket, the object name for the image in the S3 bucket, and the AWS region. The signed URL of the image is generated and returned.

Building around this, we can generate a complete solution that pulls all the available objects in the S3 bucket, and then uploads them to Roboflow via the API. An outline for this solution can be seen below:

import boto3
import requests
import urllib.parse
from botocore.config import Config

# ************* SET THESE VARIABLES *************
S3_BUCKET_NAME = "YOUR_S3_BUCKET_NAME"
ROBOFLOW_API_KEY = "YOUR_ROBOFLOW_API_KEY"
ROBOFLOW_PROJECT_NAME = "YOUR_ROBOFLOW_PROJECT_NAME"
# ***********************************************

def generate_presigned_url(bucket_name: str, object_name: str, region: str = 'us-east-2') -> str:
    """Generate a presigned URL for an S3 object."""
    s3 = boto3.client('s3', region_name=region, config=Config(signature_version='s3v4'))
    url = s3.generate_presigned_url('get_object',
                                    Params={'Bucket': bucket_name, 'Key': object_name},
                                    ExpiresIn=3600)
    return url

def get_s3_objects(bucket_name: str) -> list:
    """Fetch the list of object keys in the given S3 bucket."""
    s3 = boto3.client('s3')
    objects = []
    response = s3.list_objects_v2(Bucket=bucket_name)
    for obj in response['Contents']:
        objects.append(obj['Key'])
    return objects

def upload_to_roboflow(api_key: str, project_name: str, presigned_url: str, img_name='', split="train"):
    """Upload an image to Roboflow."""
    API_URL = "https://api.roboflow.com"
    if img_name == '':
        img_name = presigned_url.split("/")[-1]

    upload_url = "".join([
        API_URL + "/dataset/" + project_name + "/upload",
        "?api_key=" + api_key,
        "&name=" + img_name,
        "&split=" + split,
        "&image=" + urllib.parse.quote_plus(presigned_url),
    ])
    response = requests.post(upload_url)

    # Check response code
    if response.status_code == 200:
        print(f"Successfully uploaded {img_name} to {project_name}")
        return True
    else:
        print(f"Failed to upload {img_name}. Error: {response.content.decode('utf-8')}")
        return False

if __name__ == "__main__":
    # Fetch list of available images
    available_images = get_s3_objects(S3_BUCKET_NAME)
    
    # Optional: Filter the images here
    # e.g., available_images = [img for img in available_images if "some_condition"]
    
    # Upload images to Roboflow
    for image in available_images:
        presigned_url = generate_presigned_url(S3_BUCKET_NAME, image)
        upload_to_roboflow(ROBOFLOW_API_KEY, ROBOFLOW_PROJECT_NAME, presigned_url)

Option 2: Download Data Locally from AWS

Use this only when you need local preprocessing, inspection, or file transformation before upload. Otherwise, use Datasources.

To upload data from AWS, first install the awscli command line tool. This tool allows you to interact with your AWS account on the command line. Once you have the command line tool installed, run the following command:

aws s3 sync s3://mybucket/folder_path .

Replace mybucket with the name of your bucket and folder_path with the name of the folder or file you want to export. This command downloads assets from AWS to your current working directory (.).

Upload Data to Roboflow

Now that we have downloaded data, we can upload it to Roboflow either by using the Upload Web Interface or the Roboflow CLI.

See Also