Implemented a real-time class based style transfer using Fast Style Transfer and Fast SCNN-based semantic segmentation. Reduced model size to 1.64MB and improved runtime speed to 25 fps on Tesla T4 GPU.

Used Faster R-CNN to track elements in basketball games, conducting ablation studies on the model and datasets. Performed and compared various team segmentations methods.

Used IMU and image feature data to perform SLAM on a car using Extended Kalman Filter to optimize position.

Performed SLAM of a differential drive robot in a room equipped with Encoder wheels, IMU and LiDAR data, using particle filtering to improve results. Used Kinect data for texture mapping to project images captured by the robot on a map of the floor traversed by the robot.

Stabilized linear velocity data from Inertial Measurement Unit (IMU) and verified results with Vicon data. Used the resulting measurements to track camera orientation and build panorama images with projection mapping.

Built linear regression models with Sklearn in Python to identify properties responsible for toxicity of metal oxides. Studied correlation between properties to reduce dimensions and predict a numerical value of toxicity.

Built an Android app in Java to run consecutive custom interval timers useful for workouts.