Sudipta Sarkar (সুদীপ্ত সরকার)
"Education is the manifestation of the perfection already in man." - Swami Vivekananda (1863-1902)
Junior Research Fellow || Computer Science || Deep Learning || Computer Vision
Junior Research Fellow || Computer Science || Deep Learning || Computer Vision
My name is Sudipta Sarkar (Rik), a strong follower of Swami Vivekananda, Thakur Shree Ramakrishna, and Holy Ma Sharada.
Currently, I am a Junior Research Fellow (JRF) at the Department of Computer Science and Engineering, IIT Kharagpur.
Before that, I have recently completed my master's degree in Computer Science (June 2025) from
Ramakrishna Mission Residential College (Autonomous), Narendrapur,
and I earned my bachelor's degree in Computer Science (Hons.) from
Ramakrishna Mission Vivekananda Centenary College, Rahara.
I have also qualified the UGC NET (LS) examination for Assistant Professorship and Ph.D. eligibility under the June 2025 cycle.
During my five years of undergraduate and postgraduate studies, I had the opportunity to visit more than three Ramakrishna Mission Centers. I deeply admire their philosophy, work, and spiritual ideologies. Throughout this period, I met many potential friends, teachers, and monks who have continuously encouraged me to be a diligent learner and, more importantly, a good human being. I am truly grateful to all my friends, teachers, family members, and especially to the monks, whose guidance has played a crucial role in shaping me into a better person. Today, I strive not only to excel in the field of computing but also to grow spiritually and become a better individual in every aspect of life.
The name "Sudipta" (সুদীপ্ত) is of Sanskrit origin and is commonly used in Bengali-speaking regions. It is a combination of two Sanskrit words: "Su" meaning "good," "auspicious," or "prosperous," and "Dipta" meaning "bright," "shining," or "illuminated." Thus, Sudipta is often interpreted as "one who is brightly shining" (যে আলোকিতভাবে উজ্জ্বল) or "the one who is prosperous and enlightened" (যে সমৃদ্ধ ও জ্ঞানী). It conveys the idea of someone who is radiant, intellectually bright, or blessed with success.
Awarded the Best Postgraduate Student Award (among all postgraduate students across various departments) for the Academic Year 2023–2025 at Ramakrishna Mission Residential College (Autonomous), Narendrapur.
Awarded the Topper’s Medal for securing First Class First in the Master’s Degree (Postgraduate) programme in the Department of Computer Science at Ramakrishna Mission Residential College (Autonomous), Narendrapur.
Started my new position as Junior Research Fellow (JRF) in the Department of Computer Science and Engineering at IIT Kharagpur under the supervision of Prof. Abir Das.
New technical articles are now available! Explore my latest writings on Technical Article, covering Complexity Theory, Linear Algebra, Backpropagation, and Proof Methodologies.
Final Semester Thesis Presentation and submission at
RKMRC.
View Presentation PPT
Given a Seminar on Image Generation and Diffusion Models at
RKMRC.
View Seminar PPT
Started final year M.Sc. project on Video Action Recognition under Dr. Abir Das at IIT Kharagpur.
Specializing in advanced topics like Deep Learning, Computer Vision, and Algorithms. My coursework includes Machine Learning, Image Processing, and Data Science, with a focus on research-oriented projects under esteemed faculty. The college, affiliated with University of Calcutta, is renowned for its holistic education combining academic excellence with spiritual values, inspired by Swami Vivekananda’s teachings.
Studied core computer science subjects including Data Structures, Algorithms, and Database Systems. Conducted undergraduate research in Deep Learning and Computer Vision. The institution, affiliated with West Bengal State University (WBSU), emphasizes scientific rigor and ethical values.
Studied in a Bengali Medium School affiliated with the West Bengal Council of Higher Secondary Education (WBCHSE). Located in a rural area near the India-Bangladesh border with poor development and limited resources, the school faced a shortage of teachers and students. I was the only student in the science stream for classes 11th and 12th, which fostered resilience and self-reliance. Focused on Physics, Chemistry, Mathematics, and Biology, building a strong foundation in scientific principles. Participated in science fairs and regional competitions, fostering an early interest in technology.
Developing lightweight spatiotemporal models for efficient video scene parsing and action recognition. Leveraging state-of-the-art architectures including DeiT, Reversible MViT, ATLAS, Swin Transformer, Hiera, and ViT baselines. Key contributions include integration of FlashAttention-2 for faster and memory-efficient training, knowledge distillation from large pretrained models, multimodal fusion (RGB + optical flow), and SIFAR-based video-to-image reformulation.
Training ATLAS on ImageNet-1K (896×896) and ImageNet-21K (up to 1024×1024) using multi-node multi-GPU setups (8–40 A100/H100 GPUs via SLURM).
Tools: PyTorch, FlashAttention-2, einops, Vision Transformers (ViT, MViT, RevMViT, Swin, Hiera, DeiT, ATLAS, DINOv3), SIFAR Framework
Project Status: Ongoing (Junior Research Fellow position at IIT Kharagpur)
Project Link: Available soon
Supervisor: Prof. Abir Das, Associate Professor, Department of Computer Science and Engineering, IIT Kharagpur
Innovated a frame-aggregation pipeline using the SIFAR framework to recast video action recognition as image classification. Frames are rearranged via einops into super-images (e.g., 3×3 grid of 224×224 frames → 672×672 or 4×4 grid → 896×896) and processed with ImageNet-21K pretrained Hiera-ViT and ViT models.
Achieved ~79% accuracy on Kinetics-400 and ~64% on Something-Something V2 using multi-crop multi-clip testing.
Tools: PyTorch, Hiera Vision Transformer, ViT, SIFAR, einops
Project Status: Completed [June 2025] (M.Sc. Thesis)
Project Link: Available soon
Supervisor: Prof. Abir Das, Associate Professor, IIT Kharagpur
Developed a deep learning framework for stego/non-stego image classification. Extracted texture features using Gray Level Co-occurrence Matrix (GLCM), followed by pretrained ResNet50 for feature encoding, LSTM for sequence modeling, and a classification head.
Tools: PyTorch, ResNet50, LSTM, GLCM
Project Status: Completed [Nov. 2024]
Project Link: Available soon
Supervisors: Dr. Siddhartha Banerjee & Bibek Ranjan Ghosh, Ramakrishna Mission Residential College (Autonomous), Narendrapur
Engineered a compact pretrained VGG16-based model for real-time detection of 7 facial emotions (anger, disgust, fear, happiness, sadness, surprise, neutral) on the FER-2013 dataset. Emphasized quantization and low-resource optimization for potential mobile HCI deployment.
Tools: TensorFlow, VGG16, CNN
Project Status: Completed [May 2023] (B.Sc. Final Year Project)
Project Link: Show Project
Supervisors: Prof. Chayan Halder & Prof. Prasenjit Das, Ramakrishna Mission Vivekananda Centenary College, Rahara
Adapted UNet architecture for efficient segmentation of cell nuclei in biomedical microscopy images and videos. Explored potential extensions such as UNet++ and TransUNet for improved accuracy in resource-constrained settings.
Tools: TensorFlow, UNet
Project Status: Completed [March 2023] (Summer Project)
Project Link: Show Project
Supervisor: Dr. Biswajit Biswas, Ramakrishna Mission Vivekananda Centenary College, Rahara
Classifies potato leaf diseases using deep learning to improve early detection and crop cultivation in agriculture.
Project Status: Completed [Feb. 2023]
Tools: Python, Deep Learning, CNN
Project Link: Show Project
Developed a voice-interactive virtual assistant mimicking Amazon Alexa or Siri using the OpenAI API and Python libraries.
Tools: Python, OpenAI API, pyttsx3, speech_recognition, webbrowser
Project Status: Completed [Nov. 2022]
Project Link: Show Project
CLI-based game in C++ where players race to reach a target number using backtracking algorithms, designed for the machine to win if it selects first.
Tools: C++
Project Status: Completed [May 2022]
Project Link: Show Project
Contributed to an open-source collection of data structure and algorithm implementations for B.Sc. Computer Science students per UGC syllabus.
Tools: C++, Data Structures, Algorithms
Project Status: Completed [May 2022]
Project Link: Show Project
CLI-based billing system for medical stores developed in C++ during my first semester.
Tools: C++
Project Status: Completed [May 2021]
Project Link: Show Project
Project Title: Resource-Efficient Learning for Video Scene Understanding (RLV).
Under Supervision: Prof. Abir Das, Department of Computer Science and Engineering, IIT Kharagpur.
Project Title: Resource-Efficient Learning for Video Scene Understanding (RLV).
Under Supervision: Prof. Abir Das, Department of Computer Science and Engineering, IIT Kharagpur.
Managed and provided IT consulting services as part of the Vidyarthi Sabha IT Sub-Committee.
Designed and curated problem sets for Neuroverse, a college-level coding competition.
Access my detailed Curriculum Vitae to learn more about my academic and professional journey:
Feel free to reach out to me: