Name: Rupesh
Project Title: Handling Heterogeneity in Cryogenic Electron Microscopy
Academic Year: 2019-20
Semester: Autumn & Spring [7th and 8th]
Guide: Professor Ajit Rajwade
Brief Description of work: My major task was the classification of images before doing a 3d reconstruction. There were approaches based on image moments, and also using Neural Net posing the problem as supervised learning by labelling some of the images.
Area of Work: Image Processing on tomographic projections of macro-molecules
What was your motivation for this project/Area?: After doing various course projects in DeepLearning, I wanted to try projects that involved very little of Deep Learning. Also having good experience in imaging, I picked the guide first, who worked on Image processing with non-DL methods. After picking the guide, this is the project he suggested that would best suit me and I also agreed to do the same.
What was your average week like while working on the project?: We had 2 meetings per week, Monday, and Thursday in the morning time. This means I worked mostly on projects both the day before to get some results, and also just after the meeting to quickly test some of the discussions of the meeting.
What would you do differently if given a chance again? Nothing as such. Though this seems more of a philosophical question, personally my belief is that you are the sum of all your experiences. Hence, even some of the weird approaches to the problem that weren’t fruitful did give experience and learning.
How would you describe working with your guide is like?: Though having 2 meetings a week seems like a burden, but it did keep us on toes especially because as final year students, we all were mostly in a chill mode. Sir was heavily involved in our project and personally even helped me with the report. But he did expect the report to be completed a day before. So that is a really good guide to work under.
What future work do you think this project will lead to? The main task was classification and that I completed, but improving the reconstruction part still needs a lot of improvement and further there is good scope in that angle. In general, there is already some great software doing the task we were doing but we are still in the initial stage of the overall project and slow improvements can actually grow in a big project out of all this.
How difficult would you say your work was?: Work wasn’t that difficult but there was a strong skill set requirement of Advanced Image processing tools. Apart from that, basic problem-solving skills were needed and as such coming and trying the new solutions wasn’t that difficult. Some of the software was tricky and Matlab isn’t my preferred coding environment but most of the code was worked on earlier in that environment and getting familiar with the code was one of the main challenges. I even separated all the neural net parts in a different module built on python and just dumped the output to work on in the latter part of the pipeline.
Department Academic Mentorship Programme 