Make bound boxes for image with anti-captcha.com and people vision service for only 0.01$ per image
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
Readme.md
example.jpg
process.py
task.jpg

Readme.md

Make bounding boxes with anti-captcha.com crowdsourcing service. Ideal for making dataset.

anti-captcha.com is a cheap crowdsourcing service for resolving captchas and similar routine tasks.

This repository contain simple example for 'How to make bounding boxes for single class for 0.01$ per image with anti-captcha.com and people-vision fabric'.

Requirements

python3 boto and requests libs, AWS account for s3

Usage

Register on https://anti-captcha.com/ Put some money on the balance and get clientKey

lone this repository

git clone https://github.com/mnill/People-vision-example

Fullfill params in script

ANTI_CAPTCHA_CLIENT_KEY
AWS_ACCESS_KEY_ID
AWS_SECRET_ACCESS_KEY
BUCKET_NAME

Prepare two images(or use already prepared ones) and modify 'task_description' constant in file. example.jpg - Example for help workers to understand your task. task.jpg - Image to process.

run script

python3 process.py

If all works fine, you will get json with processed task, like this:

{
    'errorId': 0,
    'data': [{
        'taskId': 3139403,
        'task_description': 'Make bounded box for every tennis racquet',
        'task_image_url': 'http://s3.amazonaws.com/vision_mnill_2/task.jpg',
        'status': 'complete',
        'result_preview': 'https://vision.mnillstone.com/job/result/3139403?s=1c3c6a664375',
        'solution': [
            { 'x': 0.49298521662763,
              'y': 0.3536299765808,
              'w': 0.042868120608899,
              'h': 0.096018735362998
            }
        ]
    }]
}

For production use you may want to rewrite this code to make it parallel. There is more than 100 workers online.

License

MIT