I will build computer vision and yolo object detection models
AI Engineer for Computer Vision and Generative AI
Über diesen Service
Struggling with tracking errors or deploying a stable AI model? Most computer vision scripts break under real-world lighting, moving targets, or occlusion. I design production-grade YOLO pipelines and Face Recognition systems engineered for real-world reliability.
What I Can Build For You
- Custom Object Detection: Precision training across the entire YOLO ecosystem (YOLOv8, v9, v10, v11, etc.) to identify defects, vehicles, inventory, or PPE.
- Multi-Object Tracking: Glitch-free counting and trajectory analytics using DeepSORT, ByteTrack, or NorFair.
- Face Recognition: Secure access control and attendance systems using InsightFace or FaceNet with anti-spoofing liveness detection.
- Inference Pipelines: Optimized Python scripts for webcams, video, or RTSP streams.
What You Get
- Data Assessment: Verification of labels to prevent overfitting.
- Performance Metrics: Confusion matrices and mAP graphs.
- Clean Code: Documented modular Python scripts.
Please message me BEFORE ordering to review your dataset and requirements!
APIs:
Andere
Programmiersprache:
Python
•
SQL
•
Colab
•
MLflow
Tools:
Jupyter-Notizbuch
•
opencv
•
tensorflow
•
CVAT
•
Colab
•
PyTorch
Frameworks:
scikit-learn
•
keras
•
PyTorch
•
Panda
•
Andere
Mein Portfolio
FAQ
Do I need to provide the dataset?
Ideally, yes. If you don't have images, I can help collect and annotate them for an extra fee (please contact me first).
Can this run on my laptop/CCTV?
Yes! I optimize models (using YOLO-Nano or Small versions) to run smoothly on standard CPUs or even Raspberry Pi/Jetson devices.
Which YOLO version will you use for my project?
I work across the entire YOLO ecosystem (YOLOv8, YOLOv9, YOLOv10, YOLO11, etc.). I will select the optimal version based on your specific project needs, balancing execution speed (FPS) and detection accuracy (mAP).
What formats should my dataset be in for custom YOLO training?
I work with all standard formats, including YOLO text format, COCO JSON, Pascal VOC XML, or raw, unlabeled images. If your data isn't labeled yet, message me and we can arrange custom data annotation.
Can your systems run on edge hardware like a Raspberry Pi or Jetson Nano?
Yes. I can optimize and convert the trained YOLO models into lightweight formats like ONNX, TensorRT, or OpenVINO to ensure maximum frame rates and smooth performance on low-power edge devices.
How do you prevent face recognition systems from being fooled by photos?
I integrate software-based liveness detection using texture analysis and depth filtering. This prevents anti-spoofing, ensuring the system can tell the difference between a real human face and a digital screen or printout.
Will I get the complete source code and commercial rights?
Absolutely. Once the order is completed, you receive full commercial ownership of the clean, modular Python source code along with the trained model weights files (.pt, .onnx, etc.).

