• databases@iab-rubric.org
  • IIT Jodhpur
Df-Platter Database

Degraded and Corrupted Face Recognition Benchmark

In the proposed Degraded and Corrupted Face Recognition (DecordFace) Benchmark, we create corrupted versions of 5 popular FR datasets used for face verification, including the AgeDB (age variation), CALFW (age variation), CFP-FP (pose variation), CPLFW (pose variation) and the large-scale IJB-C datasets. A total of 80 corruptions with 16 corruption types, each at five severity levels, are utilized to corrupt the face images. In the DecordFace benchmark, we analyze the performance of over 25 popular face recognition models. These models include the popular FR models with variations in the depth of model backbones. For access to the code for the benchmark, send a request to mittal[dot]5@iitj.ac.in