Share on Facebook Tweet on Twitter Share on LinkedIn Share by email
Krishnan Ramnath

Krishnan Ramnath

We have opportunities for summer interns for 2015 in our group! See here. Candidates need to have strong programming skills in C++ (C# a bonus) and computer vision background. If interested please contact me at Summer internships at MSR are very rewarding and also a lot of fun!



I am a Research Software Development Engineer in the Interactive Visual Media Group at Microsoft Research. I have been with Microsoft Research since 2010. Prior to my current position, I was a Research Scientist at ObjectVideo Inc. from 2007. I received my M.S. in Robotics from Carnegie Mellon University in 2007. I received my B.E. in Computer Science and M.S. in Mathematics from Birla Institute of Technology and Science in India in the year 2005.


Research Interests

My research interests are in the broad area of computer vision. My current interests are in computational photography and object recognition. In my prior positions I have worked on various topics including: object detection, recognition, tracking, camera calibration and 3D modeling. Check out the projects below for a list of things I have worked on at MSR.



  • Blink

Blink is a Windows Phone app that allows users t quickly capture a burst of shots and create an animated GIF from them. The app features image stabilization to align all the photos for smooth playback as well as has automatic noise removal for enhancing those low light shots. Blink also captures a few moments before the shutter is pressed so you will never miss the moment. Blink has been downloaded 2 million times and growing. Learn more about Blink here or download for Windows Phone 8, 8.1 here.

  • IPQ: Improved Product Quantization through space partitioning

Product quantization is a popular technique for compactly encoding high dimensional vectors to facilitate fast and scalable nearest neighbor search. This is used in the context of retrieving images by query  from a large database of images. The large database needs to be encoded efficiently to fit in memory and satisfy the query speed and performance constraints. For more information see here.

  • Car Make and Model Recognition using 3D Alignment

Recognizing a car's make and model from a single image is a challenging task due to the varied orientations of cars as well as their reflecting surfaces and background clutter. In this work, we present an approach to tackling this problem by resorting to the traditional practice of building models and fitting them. We automatically generate 3D models of cars from turntable sequences and are able to recognize cars in images of cars taken in the wild. More details here.

  • AutoCaption: Automatic Caption Generation for Personal Photos

Imagine a system that automatically generates captions for your personal photos. You can make a few minor edits and send off the photo and caption to FB, Twitter or your fav social network! We developed a computer vision and language generation system that does this based on photos that are taken on a smartphone. For more about this project including demo videos, go here.

  • Automatic Text Pop-Up

The Bing homepage as interesting captions for images that are shown on the homepage in the form of Bing tiles. We have extended this idea to all the popular images on the web. For any popular image on the internet, our system pops-up interesting text snippets related to the content in the image.

We mine the internet for webpages that contain the same image and then look for interesting text to go with the image. See here for more details. Here is a link to a recent paper that uses the technology for multiple applications.


  •  Edge Foci Interest Points

Interest point detectors are widely used a first step in finding salient regions to extract descriptors for image matching. Unlike traditional detectors that compute interest points directly from image intensities, EdgeFoci uses normalized intensity edges and their orientations. We found edge foci to perform better than the existing interest point detectors such as Harris, Hessian, Laplace etc. Please see below for a demo executable that allows you extract descriptors from images.

EdgeFoci and BiCE computation module (.exe)



I am originally from Chennai, India, although, I grew up and did all of my schooling in Hyderabad, India. I studied in Niraj Public School from 3rd-10thgrade and LFJC for 11-12 grade. After that I went to BITS, Pilani for my undergrad.

During my previous job, I used to live in Reston, VA which is a very beautiful neighborhood in northern Virginia. Currently, I live with my lovely wife and beautiful daughter in Redmond, WA. I write a small blog on travels here (very sparingly updated.) Here are some pictures of us.



Please see here for a list of my publications.

Contact information:
Microsoft Corporation
One Microsoft Way
Redmond, WA 98052
Tel: 425-703-3414