Image Annotation for Emotion Recognition

The client wanted to develop a highly precise system for analyzing human emotions. He was looking for a dataset exhibiting various levels of emotional expression in individuals. Our client emphasized the need for meticulous annotation, involving over 50 points per face as per specifications. Seeking an optimal solution in terms of both price and quality for the project, the client lacked experience with various annotation tools

To ensure precision in the annotation process, we established three distinct teams tasked with annotation duties, following the client’s guidelines.

We assessed the consistency of the annotation. Cases where the consistency among all three teams fell below 98% were forwarded to the project manager for finalization. Our project manager, in collaboration with the client’s supervisor, addressed complex scenarios, accounting for 1-2% of the total volume

• The client successfully trained the neural network to recognize emotions with an accuracy rate of 99%
• Management processes were aligned with the client’s requests: a dedicated manager provided daily reports on assessor work, conducted daily meetings, and facilitated iterative revisions of the guidelines