Parcourir les catégories
Explorer
Fiverr Pro
Français
$
USD
I will build object detection, classification and pose estimation models
Computer Vision and AI Engineer, from detection and OCR to deployment
Computer-vision engineer at a production AI startup, and finalist (4th of 198 teams) of a national AI hackathon on handwritten-document recognition. I build detection, tracking and document-AI systems...
Niveau 2
Répond à des critères de performance élevés et a fait ses preuves en matière de satisfaction clients.
À propos de ce service
I train and deploy custom object detection, segmentation, classification and pose estimation models - and take them all the way to real-time production.
Real delivered examples: trading-card grading by computer vision; car-dashboard warning detection (engine, brake, AdBlue/DPF alerts) running at 15 FPS; document and label extraction pipelines.
What you get:
- Dataset strategy: annotation review, augmentation, synthetic data where it helps
- Modern models (YOLO, RT-DETR, transformers), fine-tuned on your data
- An honest evaluation report: mAP, precision/recall on a held-out set - not cherry-picked demos
- Real-time optimization: TensorRT/ONNX, GPU or edge (Jetson), FastAPI deployment
Day job: computer-vision engineer at an AI startup building real-time detection pipelines. Finalist (4th of 198 teams) of a national AI hackathon. 48 orders, 4.9 on Fiverr.
Licensing flagged upfront: AGPL models (Ultralytics) vs Apache alternatives - you choose informed.
GitHub: https://github.com/ArseniiStratiuk
Message me with your task and sample data - I'll assess feasibility honestly before you order.
APIs:
Google Cloud Vision API
Langage de programmation:
Python
•
R
•
SQL
•
Colab
•
Java
Outils:
Jupyter Notebook
•
opencv
•
tensorflow
•
Excel
•
CVAT
•
Colab
•
PyTorch
Frameworks:
Scikit-learn
•
DeepPy
•
keras
•
PyTorch
•
Panda
Mon portfolio
FAQ
How many images do I need for good results?
Minimum ~100 images per class for simple objects, 300+ for complex scenes. More data always helps, but I get the most out of small datasets with strong augmentation and parameter-efficient fine-tuning (LoRA) - adapting large pretrained models by tuning only a small fraction of their weights.
Do you provide deployment support?
I deliver models in multiple formats (PyTorch, ONNX, TensorRT) with inference scripts. Basic deployment guidance included, but complex cloud/edge deployment is an add-on service.
Can the model run in real time or on edge devices?
Yes - that's my specialty. I optimize with TensorRT/ONNX for GPU, Jetson and edge deployments, and report real FPS numbers on your target hardware, not lab benchmarks.
What about model licensing for commercial use?
I flag licensing upfront: some models (e.g., Ultralytics YOLO) are AGPL and may need a commercial license; Apache/MIT alternatives exist (e.g., RT-DETR). You choose with full information.
