01
Pose Estimation
Real-Time Body Tracking with MediaPipe
Implementation of a real-time human pose estimation pipeline using Google's MediaPipe framework, detecting 33 body landmarks at high frame rates on standard webcam input.
33 keypoints tracked in real-time via webcam feed
Custom gesture recognition logic built on top of landmarks
Angle computation between joints for motion analysis
02
Generative Models
Image-to-Image Translation with CycleGAN
Exploration of unpaired image-to-image translation using CycleGAN architecture. The model learns bidirectional mappings between two visual domains without paired training data.
Horse ↔ Zebra translation trained from scratch
Cycle-consistency loss to enforce coherent mappings
Extended to medical image style transfer experiments
Person 97%
Car 89%
Dog 74%
03
Object Detection & Segmentation
Detection & Semantic Segmentation Pipeline
Implementation and benchmarking of object detection models (YOLO, Faster R-CNN) combined with semantic segmentation approaches to understand scenes at pixel-level granularity.
YOLO v8 fine-tuned on custom dataset
Semantic segmentation with DeepLabV3+
mAP & IoU metrics tracked across experiments