0.38s Zero-Shot Vision
Type any custom class (e.g. forklift, barcode, solar panel) and get instant bounding boxes without pre-training.
8-Handle Manual Control
Resize, adjust, and drag bounding boxes with precision handles. Live normalized YOLO coordinates update in real time.
Ultralytics YOLO Ready
Export formatted .txt annotation labels, structured images/ and labels/ folders, and data.yaml.
Understanding the YOLO Bounding Box Format
In modern computer vision pipelines (such as Ultralytics YOLOv8 and YOLO11), object detection models require normalized bounding coordinates rather than raw pixel coordinates. This ensures annotations remain resolution-independent across varying input scales.
All four positional values are normalized floating-point numbers between 0.0 and 1.0, relative to the total image dimensions.
How to Train Your YOLO Model in 3 Lines of Python
Once you export your dataset archive from Polytate Studio, unzip it and run the following Python code using the official Ultralytics package:
from ultralytics import YOLO
# 1. Load pre-trained nano or medium YOLOv8 model
model = YOLO('yolov8n.pt')
# 2. Train on the Polytate exported data.yaml
results = model.train(data='polytate_dataset/data.yaml', epochs=50, imgsz=640)
# 3. Export trained weights to ONNX / CoreML / TensorRT
model.export(format='onnx')Frequently Asked Questions
Do I need to install CVAT, Docker, or LabelImg locally?
No. Polytate runs 100% in your web browser. There is no software to install, no Python environments to configure, and no Docker containers to maintain.
Are my uploaded images private and secure?
Yes. Your images and labels reside in an isolated personal workspace. We do NOT use your private customer datasets to train public models.
How many images can I annotate in Polytate?
You can upload single images or complete ZIP archives with up to 10,000–50,000 images per batch. Our parallel processing engine handles high-volume datasets smoothly.