Explorar las categorías
Explora
Fiverr Pro
Español
$
USD
I will annotate images and video for object detection and segmentation datasets
Acerca de este Servicio
Accurate bounding box and polygon annotation for computer vision training data, delivered in the format your pipeline expects.
I am a computer vision engineer, not a labeling contractor. That shows up in the edge cases: occluded objects, ambiguous class boundaries, and instances near the resolution limit. I label the way a model needs to be taught, and I flag the decisions you should make rather than quietly guessing for you.
WHAT YOU RECEIVE
Bounding boxes or polygon segmentation masks
Export in YOLO, COCO JSON, or Pascal VOC XML
Class distribution summary so you see imbalance before you train
Notes on ambiguous cases and how I resolved them
Standard and Premium include a second pass QA review of every annotation.
Premium includes a leakage safe train val test split, grouped so near duplicate frames cannot land on both sides and inflate your validation score.
Video frame extraction is included at no extra cost. Send raw footage and your sampling rate.
Message me with your image count and class list for an exact quote.
Técnica:
Manual
Tipo de etiquetado:
Imagen
•
Video
Mi porfolio
FAQ
How do you keep labeling consistent across thousands of images?
I write a short labeling guideline before starting, covering what counts as an instance, how to handle occlusion, and minimum object size, then send it to you for approval. Consistency comes from deciding the rules once, up front, not case by case.
Can you annotate video?
Yes. Send raw footage and your sampling rate and I extract frames at no extra charge. Frames from one clip are near duplicates, so I group them when splitting to keep them out of both train and validation.
Is my data kept confidential?
Yes. Your data is used only for your project, stored locally during the work, and deleted on request after delivery. I am happy to sign an NDA before you send anything.
What if I disagree with some of the labels?
Send me the specific files and what you would have done differently. I revise those and update the guideline so the same decision carries through the rest of the set. Revisions are included in every package.
