YOLO dataset

How to Annotate a YOLO Training Dataset with TjMakeBot

From label design and batch annotation to pre-export checks, this guide helps you turn YOLO dataset production into a stable workflow.

7 minYOLO2D annotationdataset export

Start by defining what the model should detect

The easiest way to break a YOLO dataset is fuzzy class boundaries. Decide what must be labeled and what should stay out of scope before annotation starts.

Write one rule for every similar class pair.
Decide how to treat occluded or blurry targets.
Use the first batch to expose rule problems early.

Use a small batch to calibrate labeling rules

Run a representative sample batch first and scale only after the rules feel stable.

01

Create the project and labels.

02

Upload an initial batch with common scenes and hard cases.

03

Let a small number of people annotate and sample-review the first round.

Check the directory structure before YOLO export goes to training

A successful export does not guarantee smooth training. Validate the images folder, labels folder, and class order first.

Open a few random label files.
Confirm class order matches the training config.
Check that both train and validation splits look correct.
dataset/
  images/
    train/
    val/
  labels/
    train/
    val/
  data.yaml

FAQ

How many images should I start with? Start with a representative sample batch, then expand volume after the rules are validated.
When should review begin? Start review with the first batch instead of waiting until everything is finished.

Suggested next learning steps

If you are new, start with annotation and export guides.
If you are preparing team workflows, continue with collaboration and plan-selection guides.
If you want the full workflow, move into OpenClaw and training guides next.
Next step

Move from content into product action

If this guide already solved the current question, use the entry points below to continue with the actual task.