Atanu Debnath (playatanu)

About

I'm a Computer Vision Engineer who loves turning research ideas into practical, real time systems. My work spans the full pipeline, data collection, model training and optimization, and deployment, from multi camera / multi IP (CCTV) video analytics for real time monitoring, detection, and video processing across many streams, to running models on Edge AI devices like NVIDIA Jetson, Rockchip, and Radxa.

I care about building efficient, reliable, and low latency solutions, and I enjoy exploring deep learning from first principles, whether that's writing a framework from scratch or squeezing more FPS out of an edge device.

Experience

Emdee Digitronics

May 2026 - Present

Computer Vision Engineer

Working on Multi IP / CCTV camera, video analytics for real time monitoring, detection, and video processing.
Developing and deploying computer vision applications on NVIDIA Jetson for real time edge inference.

Swatah

Mar 2025 - April 2026

Computer Vision Engineer

Worked on the complete computer vision pipeline, including data collection, model training, optimization, and deployment.
Optimized and deployed models using ONNX Runtime and RKNN on Rockchip and Radxa edge devices for efficient, low latency inference.

Skills & Tools

Skills

Deep Learning
Image Processing
Image Classification
Object Detection
Image Segmentation
Pose Estimation
Object Tracking
Face Detection
Face Recognition
Video Analytics
Anomaly Detection
Depth Estimation
Transfer Learning
Fine Tuning
Model Optimization
Model Quantization
Data Augmentation

Tools & Frameworks

Python
C++
OpenCV
PyTorch
ONNX / ONNX Runtime
TensorRT
TFLite / LiteRT
OpenVINO
RKNN
Jetson
CUDA / cuDNN
Docker
Git / GitHub
Linux

Projects

Pinned

In progress

The Bee Programming Language

Faster than Java and Python

Bee is a friendly programming language with a built-in native (LLVM) JIT.

BeeProgramming Language

Code

Pinned

NNRT - Deep Learning Framework

NNRT is a Deep Learning framework built from scratch in Python.

PythonDeep Learning

Code

Pinned

swatahvision - Computer Vision Library

swatahVision is a computer vision library that simplifies model loading, inference, and visualization for image classification and object detection.

PythonOpenCV

Code

simpleVision - Computer Vision Toolkit

SimpleVision is a lightweight computer vision toolkit built for fast real-time detection pipelines.

PythonPyAV

Code

Real Time Squat Counter

30+ FPS on edge

Real time workout tracking system, that detects multiple exercises, counts reps, analyzes posture, and provides live visual feedback optimized for edge devices.

Pose EstimationMovenet

Code Video

Real Time Fall Detection

Low latency

A real time fall detection system that detects sudden falls, analyzes body posture and movement, distinguishes falls from normal activities, and provides instant visual and audio alerts, optimized for low-latency edge-device deployment.

Pose EstimationMovenet

Code Video

Depth Estimation using Custom CNN (Encoder-Decoder)

A custom Convolutional Neural Network (CNN) designed for monocular depth estimation using an encoder–decoder architecture.

CNNEncoder DecoderDepth Estimation

Code Video

VIS-RPB3

VIS-RPB3 is a portable, AI-powered assistive device designed to enhance the quality of life for visually impaired individuals by providing real-time auditory information about their surroundings.

Raspberry PIOpenCVImage Detection

Code

Object Detection on Android (Flutter)

Object Detection on Android (Flutter) - A mobile application that detects and identifies objects in real time using the device camera with efficient on device processing.

YoloV8Object DetectionFlutter

Code Video

Image Similarity Search

Similarity based image search system that enables efficient retrieval of visually similar images.

ResNet50FAISSEmbedding

Code Video

Virtual Hair Colour Try-On System for Hair Makeovers

Virtual Hair Colour Try-On System for Hair Makeovers is a deep learning-based application that allows users to experiment with different hair colors on their photos.

U-NetImage Segmentation

Code Video