Case Studies Blog Careers Technologies
Synthetic Data Creation for Computer Vision Model Training

A Generative AI based solution for Synthetic Dataset Creation for specialized Computer Vision use cases.

Problem Statement

The client is working on a Computer Vision use case which has a huge need for a dataset with unique attributes - people wearing various kinds of face covers in atm rooms. Since they weren't able to come up with this data manually, they were looking for a way to automate the process of creation of this specialized dataset.


To automate the process of data creation, first we gathered the data which the client already had - people inside atm. Next, we used a Yolov5 model and trained it to detect the head of a person in an image. Using this, we created bounding boxes on the entire dataset provided by the client ie. detected the heads of people in each image and masked this portion of all these images using cv2 library.

Finally, we used Stable Diffusion’s In-painting model’s ControlNet variant to put a face cover over each image using the provided positive and negative prompts


  • Development: Python
  • Computer Vision Model: Yolov5
  • Dataset Modification: Stable Diffusion In-painting model

Value Delivered

We were able to generate images as per the client’s requirement which he was able to use to train his computer vision model