Formatting Datasets for Crimp Inspection

If you are building your own vision pipeline, standardizing how you label bounding boxes and segmentation masks is critical for interoperability with frameworks like YOLO or Detectron2.

The COCO JSON Format

We recommend exporting all annotated crimp images using the COCO format. It allows you to specify polygons for complex strand shapes rather than simple bounding boxes.

{
  "images": [
    {"id": 1, "width": 1920, "height": 1080, "file_name": "crimp_001.jpg"}
  ],
  "annotations": [
    {
      "id": 1,
      "image_id": 1,
      "category_id": 2, // 2 = Stray Strand
      "bbox": [1050, 400, 50, 120],
      "area": 6000,
      "iscrowd": 0
    }
  ],
  "categories": [
    {"id": 1, "name": "nominal_crimp"},
    {"id": 2, "name": "stray_strand"}
  ]
}