So numerous tech companies worldwide are in the race to level up and simplify this part of the Machine Learning and Deep Learning integral process. For your question answer, a good example is aerial imagery. Many startups and companies are created and running businesses with the idea of helping the planet Earth by developing applications and software that will serve for different problem-solving in various sectors (agriculture, deforestation, construction, etc).
The majority of these apps are using drone imagery or satellite photos that contain no clear lines or shapes. The development of pixel precision for aerial footage is still one of the main challenges in the image and video annotation market. Until the appearance of the appropriate software, the bounding box still remains the dominant data labeling technique, because many object detection algorithms were developed with it as a basis. Accurate pixel annotations can provide huge advantages for computer vision-based applications, but still, most tools are highly dependable on the human workforce and eligible for a human-caused mistake.
The annotator must go through every point of the photo, determine the edges of the object and label it. It is a time-consuming, sensitive and at the same time, very expensive practice. In most cases, companies are struggling to gather amounts of pixel-accurate annotations or get stuck in bounding box annotations. So we all wish that soon some software develops and upgrades enough to reduce the error possibility in object detection, and image annotation will be an undoubtedly integral part of it.