SOFTWARE ENGINEERING × MACHINE LEARNING
I'm Sharan Karuppannan. I build computer vision, learning, graphics, and computing projects, then make their ideas tangible through code and interactive experiments.
COMPUTER VISION / GRAPHICS / SYSTEMS
01 / SELECTED WORK
Research code, controlled experiments,
with workflows you can explore in the browser.
Fundus01 / 06Reference masks · workflow illustration
COMPUTER VISION · RESEARCH
Attention-WNet and Swin-WNet for retinal fundus images, with vessel-specific learning and analysis of thin vessels and bifurcations.
Live planning illustration · 3 × 4 GridWorld
REINFORCEMENT LEARNING · IIT MADRAS
DA6400 experiments in GridWorld planning, MountainCar DQN, and SAC on Pendulum. Explore value iteration in the interactive GridWorld demo.
Precomputed HOG boxes · image-space measurements
COMPUTER VISION · 2021
Detect people, capture box centers, and visualize unique pairwise distances. The portable Python entry point runs without Colab or downloaded model weights.
Canvas adaptation of the original OpenGL concept
COMPUTER GRAPHICS · 2021
Fish, bubbles, plants, and feeding scenes built with C++ and GLUT. Try the browser adaptation or build the native OpenGL project.
Trace generated by the tested Python scheduler
COMPUTER ARCHITECTURE · 2021
Step through reservation stations, dependency tags, out-of-order execution, and result broadcasts in a standalone cycle-by-cycle scheduler.
Small arithmetic illustration · native RTL is 16/32-bit
DIGITAL DESIGN · 2020
Parallel-prefix addition and structural Dadda/Wallace multiplication, with portable RTL and self-checking simulation against arithmetic references.
Live Durand–Kerner iterations · z⁴ − 1
SYSTEMS · PARALLEL COMPUTING
Durand–Kerner implementations across serial, shared-memory, distributed-memory, and GPU execution models.
Fundus01 / 04Reference mask · width measured in preview pixels
COMPUTER VISION · IMAGE PROCESSING
An OpenCV workflow for estimating vessel diameter from selected retinal vessel segments, connecting segmentation with geometric measurement.
Protocol illustration · no real IP lookup in browser
NETWORKING · REBUILT 2026
A reconstructed client/server service with bounded workers, validated JSON requests, and local IP-database lookup. An offline fixture demonstrates the protocol.
Retinal masks are dataset reference annotations, not model predictions. View image source ↗
02 / RESEARCH
Vessel-specific learning for artery–vein segmentation, including thin vessels and bifurcation regions.
03 / EXPLORATIONS
This collection spans visual perception, decision-making, computer graphics, processor scheduling, and numerical computing. Each project connects a technical idea with an implementation you can inspect or try.
Browse by topic, step through the animated workflows, and follow the source links for setup instructions and implementation details.
Python · TensorFlow/Keras · PyTorch · OpenCV · Reinforcement learning
OpenGL · C++ · Verilog · OpenMP · MPI · CUDA
GridWorld planning and continuous-control assignments.
Attention-guided segmentation of retinal vessels.
Aquarium animation, pedestrian detection, Tomasulo scheduling, and parallel root finding.
Verilog arithmetic coursework and the original TCP geolocation concept, reconstructed in 2026.
LET'S CONNECT