Ensuring Equal Distribution of Tasks Among Workers in AWS Ground Truth Private Workforce

Question

### **Description:**  
I am using **AWS Ground Truth** for an annotation task involving **bounding boxes**. The dataset consists of **Y images**, and I have **X workers** in my private workforce. I observed the task distribution is **first-come, first-served (FCFS)**, meaning that a single worker can complete **all images**, preventing fair distribution among all available workers. **Number of workers per dataset object** attribute is set to 1.

#### **What I Want to Achieve:**  
I need to **distribute Y images equally among X workers**, ensuring that each worker annotates **only a subset of images**. For example:
- If **10 images** need to be annotated and **2 workers** are available, each worker should get **5 images**.
- No single worker should complete **all** images before others get a chance.

#### **Issues Faced:**
1. **FCFS Problem:**  
   - Currently, **one active worker** can take all the images, leaving nothing for others.
   - No built-in mechanism ensures that tasks are **fairly divided**.

#### **Potential Solutions I Am Exploring:**
1. **Manually Creating Worker-Specific Manifests**  
   - Instead of a single manifest file, I create **separate manifests for each worker** and launch **individual labeling jobs**.
   - **Question:** Is this the best approach for private workforces?

#### **Key Questions:**
- What is the **best way** to **ensure equal task distribution** in AWS Ground Truth?  
- Does AWS GT **support auto-balancing tasks** among workers?  
- Can I use a **Lambda function** to filter tasks **per worker** dynamically?  
- Are there **alternative approaches** recommended for private workforces?

Alex_T · Answer

Hi, broadly you're correct: There's no built-in mechanism to limit the number or percentage of tasks each user can complete within a job as far as I'm aware... So if you want this behavior you'd need to create separate labeling jobs and assign them to single-user "teams".

You could automate this process of splitting manifests and creating/managing separate teams through Lambda, EventBridge, and similar integrations... But overall, I'd say it's a pretty unusual pattern and not one I'd particularly recommend?

Perhaps you could focus on recognition rather than hard limits: In your output manifest files I believe you should see a labeler ID, which won't be the email address of the user, but I tentatively think might be an ID you could look up against the Amazon Cognito User Pool that Ground Truth creates in your account? If you can resolve the user IDs from the output manifests to the individuals in your organization, you could produce and share leaderboards to recognize how some people contributed more to the effort than others?

This would help resolve the tension between enforcing fairness, versus prioritizing getting the work completed by whoever's available to help out.

Ensuring Equal Distribution of Tasks Among Workers in AWS Ground Truth Private Workforce

Description:

What I Want to Achieve:

Issues Faced:

Potential Solutions I Am Exploring:

Key Questions:

Relevant content