Visual QnA

This was the final project done by us for the course Neural Networks. The dataset used here is the CLEVR dataset by Stanford. The aim of this post is to introduce the reader to one of the most intriguing problems in AI. This post is not a tutorial on VQA, but a gentle intro to the problem and our approach to solving the same.

Visual QnA is one of the most challenging problems in Deep Learning. Here is the basic summary of what is Visual QnA through an example :

Image –>

Image result for CLEVR Dataset

Question 1 –>
What color is the cube that is behind the silver sphere and to the left of yellow cylinder ?

OUTPUT 1 should be :

Question 2 –>
How many  big spheres are there?

OUTPUT 2 should be :

We made a model that reached an accuracy of about 46%. This was pretty good actually, considering the best models in world have an accuracy of around 55% for numbered VQA.

Here are the links to the problem statement and our solution  :

Problem Statement

Here is the link to our model


A study of CGPA trends at BITS Goa


Birla Institute of Technology and Science Pilani, established in 1964, is one of the most esteemed engineering institutes in the country and is ranked 3rd among all engineering colleges, and 1st among private engineering colleges. The university’s second campus, KK Birla Goa campus was started in 2004 and since has laid a firm foundation for the promising future of the university. BITS conducts an entrance exam, called BITSAT, for offering admission to deserving candidates on basis of merit.

This post addresses the following issues concerning the CGPA Trends of students of BITS Goa, from various branches and courses with respect to some other parameters.

–The first graph concerns with the average CGPA of the whole batches from 2004 to 2016 with the cutoff scores of BITSAT test for Computer Science branch. In this paper, we try to gain an insight on how the average CGPA of a batch moves with the rising cutoff of BITSAT. We then try to explain the interesting outcome of the Hypothesis testing.

–Second graph concerns with the comparative performance of different branches using average CGPA and branch data

This analysis gives important insights into the sincerity of students along with the leniency of teachers in different branches.


We study how the average CGPA of a batch changes with the increasing cutoff scores. If high cutoffs imply better students, then we expect the average CGPA of that year to be more.

H1: The average CGPA of a batch with higher cutoff scores is higher than that of lower cutoff ones.


The data used in this capstone project was acquired from local IntraNet file sharing system of BITS Goa, on which it was freely available to use and share. The raw data was very diffused and had to be cleaned and sorted to get useful pieces of information. We used Microsoft Excel for this job, along with R.

The fields we get from data are:-
● Name
● ID number
● CGPA until January 2017 / passing out.
Note that the information that can be extracted from ID Number of a BITSIAN is:-
1. Year of admission
2. Engineering Branch and Science stream(for dual degree students)
3. Campus rank with BITSAT score as the benchmark.
4. Whether a student has opted for Practice School(Internship) or Thesis.

The number of entries for the data about 7,500 students; which we thought was big enough for a good project.

 Summary of Dataset





It can be clearly seen that despite rising cutoff’s, the average CGPA of batches is declining. (Possible Conclusion : level of BITSAT is getting easier every year)


Average CGPA vs Branch


A1 —> Chemical Engineering                                                   B1 —>Msc. Biology

A3 —> Electrical and Electronics Engineering                     B2 —>Msc. Chemistry

A4 —> Mechanical Engineering                                               B3 —> Msc. Maths

A7 —> Computer Sciences                                                         B5 —>Msc. Economics

A8 —> Electrictronics and Instrumentation



Yearwise Representation of CGPA of Each branch

All branches have shown constant decline except B4(Physics), which has lately seen a rise in average CGPA.

MODEL – 4 (Regression Analysis)

BITSAT Rank(ID Number) vs CGPA

It shows that if your BITSAT rank is better, You are likely to perform better than others. Since the slope is negative we can say that people with higher BITSAT scores generally score more CGPA because ID Number directly depends on BITSAT Rank and thus BITSAT Score.