Annotation Format Converter

How to use Annotation Format Converter

Convert a YOLO bounding-box dataset into COCO JSON or Pascal VOC XML when another training or evaluation workflow expects a different annotation format. Coordinate conversion changes representation; it does not improve an inaccurate box.

Prepare, run, and review

1. Create a ZIP with uniquely named images and matching YOLO text files. Use one row per box: class_id center_x center_y width height. The four coordinates are normalized relative to image width and height.
2. Use a nonnegative integer class ID and positive box sizes. Check your annotations before conversion. This tool handles five-value bounding-box rows, not segmentation polygons, pose keypoints, or arbitrary COCO input.
3. Choose COCO JSON or Pascal VOC in the output menu, then run the conversion. Image dimensions are read from the actual image files.
4. Download annotations_coco.zip or annotations_voc.zip. Open several converted examples in the receiving tool and compare them visually with the original labels.

Download a small, synthetic example dataset. It contains four labeled rectangle images and a matching copy of the first image for duplicate-check practice. It is for learning the file workflow, not for training a useful detector.

One box represented two ways

Image: 1000 × 500 pixels
YOLO row: 2 0.5 0.5 0.2 0.4
COCO bbox [left, top, width, height]: [400, 150, 200, 200]
Pascal VOC corners: xmin=400, ymin=150, xmax=600, ymax=350

The normalized center becomes 500, 250 pixels. Width and height each become 200 pixels, so subtracting half the size gives the left and top coordinates. Class IDs are retained; COCO category names are generated as class_2-style names, while VOC object names use the numeric ID. Review or map those names for your training workflow.

Limits and common mistakes

• Use unique file names across the ZIP. Output images are placed together, so duplicate names from separate folders are ambiguous and can collide.
• Unreadable images cannot be converted reliably. Run Dataset Health Check first and inspect the number of output images.
• Missing labels can produce images with no boxes. Confirm that these are intended negative examples before training.
• Different consumers use different box conventions. Verify coordinates and class mappings with the target tool instead of assuming a successful export guarantees compatibility.

Continue learning

• Validate YOLO boxes first
• Understand image coordinates

ToolsInVision guide · Updated September 26, 2026 · Editorial approach · Report an issue