This project focuses on deploying computer vision models on edge devices under tight latency and compute constraints.

Embedded AI: Real-Time Instance Segmentation

• Implemented and optimized two instance segmentation models: SparseInst and Yolov7.
• Exported models and prepared deployment flows using ONNX.
• Optimized for real-time execution on NVIDIA edge devices with CUDA TensorRT.

Practical emphasis

• Converting research models into deployable artifacts
• Latency/throughput tuning and performance validation
• Deployment-oriented engineering (runtime, memory, and stability)