Build object detection and vision systems

Design object detection/segmentation pipelines, train models, optimize for deployment

Production vision systems where both inference speed (real-time) and accuracy (>75% mAP) are constraints.

• · labeled dataset (COCO format)
• · detection requirements (FPS target, accuracy)

• · trained model (.pt or ONNX)
• · TensorRT/ONNX exports
• · mAP@50:95 metrics

• · PyTorch
• · Ultralytics YOLO
• · Detectron2
• · TensorRT

GPU recommended, Python 3.8+, dataset in COCO format

• · Small object detection fails without SAHI
• · Model overfits on dataset < 1k images

• · COCO validation script
• · mAP@50:95 benchmark
• · Quantization throughput comparison