Afnan Majeed C.

Computer Vision Engineer & Data Annotation Specialist. I build model-ready image and video datasets (bounding box, polygon, instance & semantic segmentation, keypoints) in CVAT, Label Studio, Roboflow and Labelbox, and I train the YOLO/PyTorch models they feed. ✅ WHY I'M DIFFERENT FROM A TYPICAL LABELER I've spent 5+ years as an AI Computer Vision Engineer building and deploying detection models. I know which labels improve mAP and which quietly hurt it, so your dataset is built for the model, not just for the guideline. ✅ WHAT I DELIVER • Image & video annotation: bounding boxes, polygons, instance/semantic segmentation, keypoints/pose, classification, object tracking • Tools: CVAT, Label Studio, Roboflow, Labelbox, SuperAnnotate, LabelImg • Segment Anything (SAM) + YOLO auto-labeling with human review: faster turnaround, lower cost per image • Dataset QA & audits: label-error checks, class-imbalance fixes, duplicate removal, consensus review • Any export: YOLO, COCO JSON, Pascal VOC, segmentation masks • Model training & evaluation: YOLO, PyTorch, TensorFlow, OpenCV; mAP/precision/recall reports; FastAPI + Docker deployment ✅ EXPERIENCE • Built and deployed YOLO object detection models for factory-floor visual inspection: defect detection, label verification, counting, safety monitoring. • Curated and annotated image and video datasets (bounding boxes, polygons, segmentation) in [CVAT / Label Studio / Roboflow]. Wrote labeling guidelines and ran QA to hold label accuracy. • Trained and evaluated models in PyTorch and TensorFlow with OpenCV preprocessing. Tracked mAP, precision and recall, and cut [false positives / missed detections]. • Handled real production data: lighting changes, motion blur, occlusion and class imbalance, using augmentation, hard-example mining and dataset rebalancing. • Worked with [product / operations / customer teams] to turn customer requirements into labeling specs and trained models.