SERVICES:
AI, ML, Computer Vision
Server Migration and Maintenance
Development Services
Creative Services
None
Challanges

What is the purpose of the Crop Program?

When COVID struck, the client had to start capturing images from a larger distance due to social distancing. This meant that their portrait images were now captured further back than before, and to achieve the same desired image, they needed to crop in. With the quantity of images being processed, they had to find a solution that would identify the subject and crop accordingly, ideally without user intervention. This meant they had to identify the face as well as the highest point of that human (hair, bow, etc.) and then crop around that so that the head was centered, and the perspective and aspect ratio (AR:1.5) to other images were the same.

The client needs the cropping function to run seamlessly on a remote server and to have it skip files that do not need cropping functions. The crop function should accept any size image, identify the head, and then crop to the target.

The main challenges:

  • The client wants the application to read the files from a remote server, perform the analysis, and save it on the same remote server with final cropped images in the same directory structure as the source destination without storing any images on the application server where the application runs.
  • The first productivity target was to make this a service that could run on a remote machine rather than the user's device. The end users would open a GUI, select the source folder, select the destination folder, select the functions (crop, eyes closed, glasses detection), and then submit. The service running on a remote machine should then grab this job, complete it, and ideally notify the user who submitted.
  • Multiple preset crops (V2 of this app): They will be capturing an image from a distance, and on occasions, they wanted to deliver two different crops back to the destination folder.
  • Investigate whether green screen removal (chroma key) could be actioned at the same time as the crop function.
  • Create all APIs of this application so that the client can use this anywhere in any application.
  • Clients also want to crop 50,000 images from one server in a day without using GPU's Server.
Solutions

We tackled the challenges of this project head-on, achieving a 100% completion rate with an accuracy of 99.99%. Leveraging machine learning, computer vision, parallel processing, and multitasking, we surpassed the target of processing over 50,000 images per day, all while reducing the need for human intervention.

Our solution seamlessly integrates with remote servers, ensuring that the application can be deployed across multiple branches.

This application not only enhances business profitability but also streamlines task completion, eliminating concerns about accuracy and precision.

None
None
None
None
None
(011) 6138-1949
Mon - Fri 10:00 AM - 06:30 PM