Intern Diaries : Indian Statistical Institute, Kolkata
Smruti Sucharita Nath | Mar 16, 2020
The Indian Statistical Institute, Kolkata is a unique institution devoted to research and teaching, and the applications of Statistics in the Natural and Social Sciences. The institute conducts summer school programs every year in the field of computer vision and image processing. 6 of the students, from circuital branches, enrolled for this internship after their 4th semester in the summer of 2019.
Eligibility: Undergraduate students (2nd year onwards) pursuing their degree in circuital (Computer Science, Electrical, Electronics and Communication/Instrumentation) branches.
Research areas: Computer vision, Computer Graphics, Data science, Image Processing.
Time of application: The application for the internship usually begins during Mid-April.
Application Procedures: One needs to fill an online application in the portal, www.isical.ac.in. The candidates need to have a brief resume, letter of recommendation, permission from the Head of the Department of the Institute. Shortlisted candidates get informed through email.
Accommodation: Accommodation is given to those who apply early and an amount of RS. 100 is charged per day. The canteen at ISI provides good meals at very cheap rates one can have his lunch within Rs.10.
BISWAJEET NAYAK (3RD YEAR, DEPARTMENT OF ELECTRONICS AND INSTRUMENTATION)
I came across the program from some club seniors who did their internship there during the previous year. The shortlisting was resume based primarily. So having some projects on programming or knowing the basics of machine learning/image processing would suffice as the prerequisites. We initially proposed to work on GAIT, but our mentor denied this, stating that it would be something difficult for us. So he suggested we work on action video classification, and we did a bit of looking into the project and took it up.
The academia was inspiring. The computational resources were state-of-the-art, consisting of graphics enabled systems costing up to lakhs. We were initially not allowed to use the systems as the training phase started. We were not able to continue the process because our laptops were not graphics enabled, so we were allowed to use the lab computers then.
Our project included classifying various videos into different classes. We used the UCF-101 dataset, which included 101 action classes with around 90-10 videos in each class. We used CNN-RNN combined networks along with a fusion technique to classify the videos. Later we added an IoT module to it so that the whole system becomes smart enough to detect an anomaly video class(any crime or something) and send the user notification about the activities happening.
The project which we undertook was somewhat complicated for us at that stage. Also, the data we got after extraction came out in GB’s and when we trained the model initially, we got an accuracy of 8-9%, so our mentor asked us to identify and sort out the problem and improve the model. We struggled with the problem for more than ten days. Finally, our mentor helped us in detecting the problem just before the presentation day, and the major progress of our project was done on that day.
I learned many things that helped me during this academic year and while taking up new projects. I somewhere heard that ML models start with research papers and end with presentations. Through this internship, I went ahead of the pipeline and deployed a model in the real world. We are recently also about, to begin with, the project again and finish it up. ISI internship was my first foray into the field of research. Despite the fact that I plan to dive into the corporate world soon enough, but this experience will always be cherished by me.
AMARTYAA AVIZEETA (3RD YEAR, DEPARTMENT OF ELECTRICAL ENGINEERING)
I had worked in the field of Image Processing and Deep Learning before this internship and gained interest in how these fields could be integrated with automation. I focused on Computer Vision and Deep Learning. My project was based on Semantic Segmentation using the concepts of Computer Vision and Neural Networks. We trained a model such that, when some random image was fed into it, it would classify the objects in the image into the predefined classes and also give the exact spatial location of the classes in the image. They completed every topic in depth. The schedule was divided as, few days of theory and hands-on class followed by practical implementation through projects. PhD students at the institute are assigned as guides for the teams while working on projects. The working environment was pretty competitive too because, in the end, some projects received special mention.
The stay at the institute was quite comfortable. The project that we did helped us to gain a lot of knowledge in the field. I feel doing a project or practically implementing the concepts proves to be more useful than just going over the concepts again and again. Since your CV carries a lot of weight in the selection procedure, it's advisable to have some prior knowledge or some kind of similar project to that of the fields of the internship. This summer internship proves to be very useful for the students who aspire to build their career in Computer Vision and Data Science. And since Data Science is the most emerging field these days, such an internship is going to give you a lot of knowledge in the field.
Team Monday Morning wishes all the aspirants and the former interns a stroke of good luck for their planned endeavours.