Dr. Abhishek Dutta

Archive for the ‘PhD Research’ Category

“Abhishek Dutta’s research”

It is never easy to explain your research project to a general audience. At several occasions, I have found myself in the middle of a conversation when I am explaining my work and I see a poker face in my humble and patient listener. It is then that I realize, how bad I am at explaining my work. Every such “poker face” helps me to improve my ability to reach to a general audience.

So I was a bit nervous when I learned that I had to give a skype interview on Dec. 22, 2011 to a professional content writer for my research group’s flyer (a leaflet intended to publicize ongoing research projects). I scribbled a few important points in a paper and started my interview with some courage obtained by believing that I had already seen enough “poker faces” by now. The interviewer was very smart and quickly understood my work and its importance — if any. :)

The leaflet has now been published (flip side contains my supervisor’s interview) and I am very happy with its content. I lost no time in reading it and quickly snail mailed it to my parents after I received them in my mailbox. Now here I am, pompously talking about it in my blog. To be honest, it feels very nice when somebody writes about your research. Also, I am a Leo and nothing flatters me more than the very though of being talked about — of course, in a good way. So when I first read Linda Goodman’s description of a Leo (about 4 years ago), it felt as if she was writing about me: all my traits (good and bad) were laid naked for the world to see. Felt really embarrassed but again in a good way: I did not choose to be a Leo. :)

Written by abhishekdutta

June 7, 2012 at 8:13 pm

Posted in PhD Research

Frustrations from the current state-of-the-art in automatic face recognition.

For the past few weeks, I have been carrying out some forensic face recognition experiments based on the CMU MultiPIE dataset. The objective of these experiments is to test a new framework for forensic face recognition. Therefore, in these experiments, I use a commercial face recognition system as a “black box”. In other words, in these experiments, I am not concerned with how this commercial face recognition system works but I aim to present a framework for applying automatic face recognition systems to forensic cases.

Today, I finished the last part of the experiment and I feel so frustrated from the results that I have decided to express my frustrations in this blog post. To be honest, I had never been so frustrated in the past few years.

I have the following appeal to all the researchers (including myself) involved in face recognition:

  1. Try to venture outside the comfort zone of controlled face image dataset. Most face recognition algorithm developers feel safe to test their algorithms using only standard face image dataset captured in controlled (pose, illumination, etc) environment. Please have the courage to test algorithms on uncontrolled images and face the reality.
  2. Stop pursuing research to improve face recognition accuracy from 96% to 98% on a controlled face image dataset. We have already lost the past decade doing such incremental research.
  3. Stop using idiosyncratically chosen facial features which can hardly be attributed to an individual’s identity. Such rouge facial features are often found wandering in the feature space when subject to facial pose and illumination variation.

It is not difficult to understand why the professionals involved in forensic face recognition cases prefer to use trained human experts for face matching. Today’s frustration has a bright side as well. I now feel happy to be exploring a dilapidated research avenue. Not only more reasons to be frustrated but also more reasons to try out new ideas and challenge the conventional wisdom.

Written by abhishekdutta

August 16, 2011 at 9:19 pm

Posted in PhD Research

PhD Research Activity Level

PhD Year 1 Activity (Last Updated: 29 Jul 2011)

Caution: This plot is just for fun and should not be taken seriously. I did it when I got bored with my research one weekend :D

I have meeting with my supervisors after every two weeks (bi-monthly meetings). I email the meeting agenda to my supervisors before every meeting. After finishing the meeting, I again email the meeting minutes (summary of important discussions during the meeting) to my supervisors. I have done this for last 6 months and I suddenly realised that I can use this data to assess the research activity level during my PhD.

The agenda and minute of each meeting has a set of bullet points. The number of bullet points corresponding to agenda and minute of each meeting represents the level of my research activity in the past two weeks before that meeting date. If I am very active and working hard on some research topic, the next meeting is bound to have a lot of bullet points corresponding to related discussion during the meeting. I use this count of meeting activity as a way to estimate my research activity. It is interesting to see that these plots somehow succeeded to depict the amount of work that I have been doing during my PhD research.

This way of measuring research activity has some serious limitations:

  • Not all bullet points in agenda and minute have same weight. Some points in agenda can take about one hour to discuss while others can be resolved in just few minutes. Therefore, it is not fair to attribute equal contribution of each bullet point towards overall research activity level.
  • Serious discussion on a topic during meetings requires very good understanding of different aspects of that research topic. Developing such a level of understanding can take several weeks. Therefore, the underlying assumption that activity during a meeting is a result of work done in the past two weeks is not always true.

In spite of these limitations, I decided to create this plot because it gives a good (although not very accurate) overview of my PhD research activity. I am waiting for the last meeting at the end of my PhD and wondering how such a plot for 4 years would look like.

Written by abhishekdutta

July 29, 2011 at 7:03 pm