Performed Multi-class product counting and recognition from real-time videos for automated retail checkout with an F1-score over 45%
Built a cascading pipeline consisting of a module foreign object identifier, a hand segmentation module using U-Net, an entropy based domain bias detector and Vision Transfomer based classifier.
Developed a custom heuristic-based filtration metric to discard empty frames and improve precision.
Selected as one of 66 interns from a pool of 4500+ applicants. Under the supervision of Dr. Isaac Johnson and Dr. Martin Gerlach, I worked with the Wikimedia Research team for three months to build a parser for Wikimedia Enterprise HTML dumps. Check out the project blog for more details.
I joined the 2021’s GHC event, as a student scholar. Less than 5% applicants are chosen as scholars each year for their extraordinary academic history and future potentials. I was fortunate to be able join the pool of hand-picked scholars in my first participation.
With our ML-based method to quickly diagnose the symptoms of various neonatal diseases and STDs, we secured the 3rd place in the regional finals of the Microsoft Imagine Cup.
2020
4th place in DhakaAI - AI-Based Dhaka Traffic Detection Challenge
Secured 4th place out of 250+ teams in this international Computer Vision challenge organized by DhakaAI.
Cleaned and pre-processed the noisy, poor quality dataset to fit with different light and weather conditions through augmentation and manual data collection.
Tuned the hyperparameters of 5-8 SOTA Image classification models (EffDet, Yolo, Cascaded-RCNN, Faster-RCNN) and applied ensemble technique with WBF.
My team developed a prototype android app that would facilitate the ASD (Autism Spectrum Disorder) affected by improving their overall social and communication skills. The project mainly utilized VR and AR technologies to provide a fun and interactive learning environment. The project was awarded the best project in the hackathon beating over 50 teams from different universities of Bangladesh.
My team became champion competing with over 300 teams from all over the country across various domains. We built a system that contained an app and web dashboard of learning & assessment modules for BRAC’s 6000+ Health workers. See the <a href="https://blog.brac.net/building-capacity-of-community-health-workers-remotely/> project page</a> for more details.
