Browse categories
Explore
Fiverr Pro
English
$
USD
I will annotate and label your image dataset for yolo object detection
Builder and self taught engineer
I build systems that address real-world problems and pain points.
My work spans software, electronics, and mechanical design.
https://markosnarinian.com
About this Gig
I train computer vision models - that is why my annotations are different.
I built a road damage detection system end to end: collection, annotation, training, edge inference on Raspberry Pi, benchmarking, and the cloud pipeline behind it. The lesson was expensive: a model is only as good as its labels.
Every project includes a written class spec agreed before labeling, explicit edge-case rules (occlusion, truncation, min size), a second-pass QA review, a class distribution report, and clean train/val/test splits. Formats: COCO, YOLO, Pascal VOC, CVAT, CSV.
Boxes, polygons, keypoints, classification, video tracking.
I will tell you if your dataset won't train BEFORE you pay, not after.
Premium delivers the dataset plus a trained, benchmarked model with an edge-ready export.
Send a few sample images and tell me what you need detected.
Technique:
Automated
Tagging type:
Image
•
Video
FAQ
How is this different from the $5 annotation gigs?
Those sellers deliver boxes. I deliver a dataset that trains. I annotate against how the model learns: consistent class definitions, resolved edge cases, checked class balance, and a QA pass. A model that plateaus for no reason usually has a labeling problem.
Can you tell me if my dataset is good enough before I order?
Yes, and I'd rather do that first. Send 10-20 sample images and your class list. I'll tell you honestly whether the taxonomy is workable, roughly how many images you need, and what the failure modes will be. No charge for that conversation.
What formats do you deliver?
COCO JSON, YOLO (txt), Pascal VOC (XML), CVAT XML, and CSV. Need something else? Just ask, conversion is straightforward. Standard and Premium include multiple formats plus train/val/test splits.
What tools do you use?
CVAT, Label Studio, LabelImg and Roboflow depending on the project. If your team already works in a specific tool and can give me access, I'll work directly in yours so the data never leaves your pipeline.
How do you handle ambiguous or edge cases?
I surface them instead of guessing. Before labeling starts we agree a written spec covering occlusion, truncation, minimum size, and borderline cases. Anything new that comes up mid-batch, I send you examples and we decide together.
Can you also train the model?
Yes, that's the Premium package. You get the annotated dataset plus a trained detector, a benchmark report with mAP and per-class metrics, and an export ready for edge or server deployment. Tested on real Raspberry Pi hardware.
Is my data confidential?
Yes. I don't reuse, resell, or publish client data, and I'm happy to sign an NDA before you send anything. If your data can't leave your infrastructure, I'll work inside your own annotation tool.
What do you need from me to start?
Your images or video, a list of the classes you want detected, and a sentence on what the model is for. That last part matters more than people expect: labeling for a safety-critical detector is a different job than a rough dashboard.
How fast do you reply?
Usually within the hour during European working hours (I'm in Greece, EET/EEST). I work in English and Greek.
Why is the Pilot package so cheap?
I'm building my track record here, and I'd rather earn it with real work than ask you to gamble on someone with no reviews. The Pilot tier is priced as an introduction, not as what the work is worth - it goes up once I have a few completed projects. Same spec and QA as every tier.
