Selected work
Computer vision
The Tri-Stage Occlusion Handling Normal Map Estimation Algorithm generates accurate 3D normal maps from heavily occluded 2D images by handling background, shadow, blur, depth, and orientation cues.
92.78% IoU, compared with a 71.26% baseline.
View codeEdge AI
A static quantization approach for AlexNet, MobileNet, and ResNet-50 models trained on retina-based diabetes classification data, reducing model size while preserving performance.
View codeNatural language processing
A PyTorch BERT-based sequence classifier for distinguishing paragraphs by whether their content remains evergreen or ephemeral over time.
View codeWearable computing
A mechanical exoskeleton and learning pipeline for translating American Sign Language gestures into English using finger-joint sensor measurements.