This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
public:gsoc:poormanrekognition2 [2021/02/13 19:45]
public:gsoc:poormanrekognition2 [2021/03/14 21:07] (current)
Line 3: Line 3:
 [[|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.  [[|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. 
-This is a continuation of a project that has been running the past 2 years: The first year we had a number of students working on this in parallelwhich in itself was an experiment we wanted to see if having several students in parallel try to do their own implementation would be better than just choosing the one we thought would best. Last year we selected only one, continuing on the best project of the first year+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 sceneand 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.
-The repository can be found here for PMR-III:
-This year the goal would be to extend and build upon the existing software, adding features and fixing existing bugs.+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.
-**Qualification tasks**\\+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.
-- Take a look at [[|this page]] for more tasks.+Deploy the service for remote useOnce the above two issues have been resolved 
-**Older (but might still be relevant!) links**\\+Domain: Artificial Intelligence, Deep Learning, Computer Vision
-[[public:gsoc:poormanrekognition|This]] is the description to last year's project.+**Relevant links**
-\\ +[[|Source Code]]
-[[|Setting Up the Project and Brief Overview]]
-**In detail blogs**
-[[|GSoC Chronicles — Only Time will Tell]]\\ 
- +[[|]]\\ 
-[[|GSoC Chronicles — commit the CRNN cometh the Text]]\\ 
 +[[| ]]
  • public/gsoc/poormanrekognition2.1613245515.txt.gz
  • Last modified: 2021/03/14 21:07
  • (external edit)