Our team "Kakatua" became 1st Runner-up and got an award of tk. 50,000

The main objective of this competition was to develop a system that can transcribe Bengali speech, including different regional dialects, in accordance with the orthographic standards established by linguists. Our team “Kakatua” worked on fine tuning wav2vec2, whisper and conformer models.

For details of the competition click

My notebook became the best notebook and got an award of tk. 15,000

This is a notebook for beginners who are inerested to work for Bengali NLP. The notebook shows basic normalization, error analysis for provided data and simple model training using bangla-T5

For more details click

Our team "Sherlock" became 2nd Runner-up and got an award of $1500

I was a mentor of the team

The challenge of this competition was Detecting Synthetic Images. In our work, a deep learning based binary classification pipeline has been proposed for classification of real and Synthetic or partially manipulated images have been proposed. In our approach, features extracted from both from the image and its Fourier transformed version along with various augmentation such as cut-mix and flipping has been used.

For relevant code click

For details of VIP Cup click

Our team "Ekush"  became 1st Runner-up and got an award of tk. 60,000

The main task of the competition was "Bengali Speech to Text Conversion". We were provided a rich audio dataset of 200h+ with Bengali encryption. We have used a wav2vec model to solve the problem with proper hyper parameter tuning. Later we have published the model in Hugging Face.

For relevant code click

For details of the competition click

Our team "Samaritan" was in top 10 among 100+ teams from all over the country.

I was a key member of the team.

The main challenge of this competition was to handle big data. We were given a large dataset of about 2 crore transaction between customers and retailers from all over the country. We have analyzed and found answer of 11 analytical questions.

For details of the competition click

Our team "Samaritan"  became Champion and got an award of $5000

I was a key member of the team and got a "Token of appreciation" from "Directorate of Student's Welfare, BUET" for this work.

The main challenge of the competition was "In-bed Sleeping Pose estimation". The organizers provided us a subset of the SLP dataset that contains IR images of subjects lying on the bed on different poses. In our work, we developed a deep learning based framework for human sleeping pose estimation from infrared images.

For relevant code click

For details of competition click