public:gsoc:poormanrekognition2

This is an old revision of the document!


Amazon Rekognition is a (paid) service that is able to identify objects, people, text, scenes, and activities in a picture. We want to produce a free alternative.

Description: Rekognition is Amazon’s (paid) service capable of identifying objects, text and activities, performing facial analysis and recognition, detecting the frequency of objects or an inappropriate scene, and much more using deep learning. Poor Man’s Rekognition is an open-source version of the commercial service and is currently able to do almost everything that Amazon Rekognition does.

Ideas for PMR-III:

Fix issues with docker and make it run seamlessly on all platforms: The project as of now runs specifically on Unix based distros. The docker image is broken and needs to be fixed. The end goal is to have a containerized version of the project that can run on all platforms.

Decrease latency of API response by reducing the size of some of the large models (or by some other means): Some of the models used in PMR are extremely large in size and thus take a lot of time to get loaded and give predictions. The memory consumption is also extremely high. We need to decrease both the response time as well as memory consumption. One way of doing this can be using models with smaller size. Other methods can be explored.

Deploy the service for remote use: Once the above two issues have been resolved

Domain: Artificial Intelligence, Deep Learning, Computer Vision

Relevant links

Source Code

Setting Up the Project and Brief Overview

In detail blogs

GSoC Chronicles — Only Time will Tell
https://medium.com/@pulkitmishra/gsoc-chronicles-best-kept-code-4893823d0f12
GSoC Chronicles — commit the CRNN cometh the Text
https://medium.com/@pulkitmishra/gsoc-chronicles-mightier-than-ssd-b89be236d852

  • public/gsoc/poormanrekognition2.1613666157.txt.gz
  • Last modified: 2021/03/14 21:07
  • (external edit)