Annotation Best Practices for Object Detection

---

1. Tight Bounding Boxes

The boxes should be as small as possible to fit the entire object

2. Completeness

Make sure you have annotated all objects from all categories in an image. Missing any objects hampers the model’s learning

3. Consistency across images

You should maintain uniformity in annotations across all images in case your object has a certain degree of subjectivity in identifying them. For instance, how damaged the concrete surface should be to call it a “crack” must be the same for all images