Quick Answer

YOLO object detection labels usually store one .txt file for each image. Each line describes one object. The common format is:

class_id center_x center_y width height

The four box values are normalized. That means they are not pixel values. They are ratios from 0 to 1 relative to the image width and height.

YOLO label format diagram with bounding box, center point, width and height arrows, and abstract label rows

The image shows the idea. The object is surrounded by a bounding box. The label stores the class and the box position using the center point, width, and height.

Basic Example

Suppose an image contains one bottle. The label file might contain:

0 0.500000 0.520000 0.250000 0.640000

Read it like this:

Value Meaning
0 class id
0.500000 center x position
0.520000 center y position
0.250000 bounding box width
0.640000 bounding box height

The class id is an index. If your class list is:

0 bottle
1 cup
2 box

Then class id 0 means bottle.

One Image, One Label File

In a YOLO dataset, image and label files usually share the same base name.

images/train/photo_001.jpg
labels/train/photo_001.txt

If photo_001.jpg contains three objects, photo_001.txt has three lines. If the image has no objects, the label file may be empty or absent depending on the training pipeline. Always check the expected behavior of the tool you use.

Why Values Are Normalized

YOLO uses normalized coordinates so labels remain valid when images are resized during training. For example, center_x = 0.5 means the center of the image horizontally. It does not matter whether the image is 640x480, 1280x720, or another size.

The conversion is:

center_x = box_center_x_in_pixels / image_width
center_y = box_center_y_in_pixels / image_height
width    = box_width_in_pixels / image_width
height   = box_height_in_pixels / image_height

To convert back to pixels:

box_width_pixels  = width * image_width
box_height_pixels = height * image_height

This is the most common place where labeling mistakes happen. People sometimes write pixel values into the YOLO text file. That usually breaks training because the values should be between 0 and 1.

Center Format, Not Corner Format

YOLO labels use center format. Many annotation tools and image libraries use corner format:

x_min y_min x_max y_max

To convert corner coordinates to YOLO format:

box_width  = x_max - x_min
box_height = y_max - y_min
center_x   = x_min + box_width / 2
center_y   = y_min + box_height / 2

Then normalize those values by image width and height.

If you convert from COCO, Pascal VOC, or custom CSV annotations, check this carefully. Wrong coordinate conversion can make boxes appear shifted, too large, or too small.

Class IDs Must Match the Class List

The label file stores only the numeric class id. The class name usually lives in a dataset configuration file or class list.

For example:

0 car
1 bus
2 truck

If you later reorder the class list, old labels can become wrong. A label that used to mean car may suddenly mean bus.

Use this rule:

Once labeling starts, do not reorder class IDs.
Add new classes at the end when possible.

This is especially important when multiple people label data together.

Common YOLO Label Mistakes

The first mistake is using pixel coordinates instead of normalized coordinates. If you see values like 340 220 120 80, that is probably not valid YOLO format.

The second mistake is using top-left corner coordinates instead of center coordinates. YOLO expects the center of the box, not the top-left corner.

The third mistake is mixing image and label filenames. photo_001.jpg should match photo_001.txt.

The fourth mistake is class id drift. This happens when the class list changes after labels were created.

The fifth mistake is drawing loose boxes. Object detection labels should be tight enough to describe the visible object, but not so tight that important edges are cut off.

Verification Checklist

Before training, sample several images and check:

[ ] Every image has the expected label file.
[ ] Every object line has five values.
[ ] Class IDs exist in the class list.
[ ] Coordinate values are between 0 and 1.
[ ] Boxes appear over the correct objects.
[ ] Boxes are not shifted after resizing.
[ ] Empty images are handled consistently.

Visual inspection is important. A dataset can look valid as text and still have boxes in the wrong place.

Easy Labeling Workflow

If you use Easy Labeling, the key workflow is:

1. Load an image folder.
2. Load or define the class list.
3. Draw boxes around objects.
4. Save labels in YOLO format.
5. Reopen a sample image and verify the boxes.

The verification step is worth doing before you label hundreds of images. It catches class-order mistakes and coordinate export problems early.

You can try the tool here: Easy Labeling.

Easy Labeling Screen Example

The screen below shows the basic flow: open an image, draw a box, and assign a class.

Easy Labeling sample screen for drawing object detection boxes

Final Advice

YOLO label format is simple, but it is strict. One line means one object. The order is class id, center x, center y, width, height. The box values are normalized, not pixels.

If you remember those rules and verify visually before training, you avoid many object detection dataset problems.

FAQ

When should I use this guide?

Use it when you need a repeatable labeling workflow, cleaner dataset handoff, or clearer review rules for image annotation.

What should beginners verify first?

Start with class definitions, positive and negative examples, review criteria, and export format. The tool works best when the labeling rule is explicit.

Which keywords should I search next?

Search for “YOLO Label Format: How to Read Class, Center X, Center Y, Width, and Height” together with image labeling, dataset annotation, YOLO, COCO, review workflow, and labeling quality keywords.

Leave a comment

You may also enjoy