I will train a custom yolo object detection model and deploy it as an api
About this Gig
Need software that can spot, count or track objects in your images or video? Products on a shelf, vehicles at a gate, defects on a line, PPE on a site? I will train a custom YOLO object detection model on your data and deliver it ready to use.
What you get:
- Dataset review: I check your images and labels and fix problems before training
- Custom YOLO training for your classes, with augmentation and tuning
- An evaluation report: precision, recall and mAP per class, plus examples where the model fails
- Inference on images, video files or live camera streams
- Optional counting, tracking and zone alerts with OpenCV
- A FastAPI endpoint in Docker, so your app can send an image and get detections back
- Model weights, ONNX export and documented source code
Stack: Python, PyTorch, YOLO, OpenCV, MediaPipe, FastAPI, Docker.
I am an AI engineer with a data science background and 25+ freelance AI and machine learning projects behind me.
Results depend on your data, so please send 20 example images or a short video before ordering. I will tell you honestly whether detection is feasible and which package fits.
I do not take academic assignments or exams.
Expertise:
Image processing
•
Classification
•
Object detection
Programming language:
Python
Tools:
Jupyter Notebook
•
OpenCV
•
TensorFlow
•
CVAT
•
PyTorch
Frameworks:
Scikit-learn
•
Keras
•
PyTorch
FAQ
What is included?
A custom-trained YOLO model, an evaluation report with per-class metrics and failure examples, inference code and the model weights. Standard and Premium add tracking, an API and deployment.
What is not included?
Camera hardware, cloud GPU costs, and labelling in Basic and Standard. I do not build facial recognition of identifiable people or do academic assignments.
Do I need labelled data?
For Basic and Standard, yes (YOLO or COCO format). Premium includes labelling up to 800 images. Not sure? Send samples and I will advise.
Can it run on my existing system or camera?
Usually yes. The FastAPI endpoint accepts images from any app, and I can read most IP cameras via RTSP. Tell me your hardware first.
How accurate will it be?
That depends on image quality, variety and labels. I report precision, recall and mAP honestly and show where the model fails, so you can decide before going live.
What access do you need?
A read-only dataset link and, for deployment, a temporary SSH key on your server. No passwords in chat.
What if I need more classes or features later?
I can retrain with new classes or add features through a custom offer. Your source code and weights stay yours.

