Responsibilities
- Prepare and clean drone image segmentation datasets
- Assist with model training and evaluation
- Extract meaningful features from low-quality data
- Improve & maintain data creation & consumption pipelines
- Create quality annotations manually & automatically
- Prioritize the data that will make the best impact
Must have
- Python (numpy, OpenCV, JSON/CLI scripts)
- PyTorch basics — train/fine-tune a model
- Detection vs. instance vs. semantic segmentation; COCO & YOLO formats
- Understanding of metrics (mAP, IoU) + training curves
- Comfortable with messy data; Git, Linux
Nice to have
- CVAT / SAM annotation, Ultralytics YOLO, DETR/HF transformers
- Augmentation + fine-tuning intuition, cloud GPU (AWS)
- Curious and passionate about data
- Careful with data: understands "garbage labels → garbage model"