Computer Vision
project

Accelerating the performance of the object detection project

This case study shows our project related to image processing and computer vision. The main goal was to achieve significant performance acceleration of existing solution.

Industry:

Computer Vision

Country:

Norway

Timespan:

2021/2022

Link to the project:

Project’s tech stack

Thanks to the applied changes, the application has been accelerated by 20 times.

Project & Client

Client is under strict NDA.

Our goal was to speed the application (more specifically the Mask R-CNN's inference) so that the processing of high-resolution photos would be faster and more efficient. At the same time, it was crucial not to deteriorate the application's capabilities.

Solution

We approached accelerating the Computer Vision project in three key steps:

  1. We moved computation from CPU to GPU.
  2. We optimized the mixing of blurred and the original image
  3. We switched the Gaussian filter to the Box Filter.

Before our improvements, the application needed ~3 minutes to process one image, and the inference of a neural network occupied 70% of this time. After changes made by Flyps’ experts, the time it took for neural networks decreased to 9 seconds. Thanks to the applied changes, the application has been accelerated by 20 times.

Want to know
full details of the project?
Check our blogpost here

close