At the 2021 edition of the Computer Vision and Pattern Recognition (CVPR) Conference, Siddhartha Bal, Assistant GM of Delivery and Operations at iMerit presented a talk entitled ‘The Human Brain’s Ability to Create ‘Context” for Improving Data Annotation Accuracy’. Siddhartha and his team have annotated over 100 million images for computer vision projects and are considered experts in some of the most complex use cases in the industry. He discussed some of the unique data challenges iMerit has come across while working with clients and how our annotation team leveraged context to develop effective solutions.
Here are three key takeaways from the session:
- Annotation is not a binary task. Given that each human mind thinks differently depending on their experiences, providing annotators with all contexts concerning the images will allow them to make the right judgments during complex labeling situations, resulting in improved annotation accuracy.
- It is important to ensure that images of the same sequence are provided to one particular annotator. If different labelers mark different frames of the same image, the annotations will be out of context, resulting in inconsistent labeling.
- Sharing the evaluation mechanism with the annotators improves context and visibility and ensures that the output meets the client’s expectations.
To watch the on-demand webinar, click here.
