Achieving Better Kinship Recognition Through Better Baseline
This work addresses kinship recognition for applications like family search, but it is incremental as it builds on existing face verification methods.
The paper tackled kinship recognition by transferring knowledge from face verification to create a new baseline, achieving state-of-the-art performance on two tracks in the Recognizing Families In the Wild Data Challenge.
Recognizing blood relations using face images can be seen as an application of face recognition systems with additional restrictions. These restrictions proved to be difficult to deal with, however, recent advancements in face verification show that there is still much to gain using more data and novel ideas. As a result face recognition is a great source domain from which we can transfer the knowledge to get better performance in kinship recognition as a source domain. We present a new baseline for an automatic kinship recognition task and relatives search based on RetinaFace[1] for face registration and ArcFace[2] face verification model. With the approach described above as the foundation, we constructed a pipeline that achieved state-of-the-art performance on two tracks in the recent Recognizing Families In the Wild Data Challenge.