Quad-Channel Contrastive Prototype Networks for Open-Set Recognition in Domain-Specific Tasks

A traditional deep neural network-based classifier assumes that only training classes appear during testing in closed-world settings. In most real-world applications, an open-set environment is more realistic than a conventional approach where unseen are potentially present the model’s lifetime. Open-set recognition (OSR) provides model with capability to address this issue by reducing risk, which unknown could be recognized as known classes. Unfortunately, many proposed techniques evaluate performance using “toy” datasets and do not consider transfer learning, has become common practice deriving strong from learning models. We propose quad-channel contrastive prototype network (QC-CPN) views of input loss for applications. These also require tuning new hyperparameters justify their performance, so we first employ evolutionary simulated annealing (EvoSA) find good our approach. The comparison results show QC-CPN effectively outperforms other state-of-the-art rejecting domain-specific dataset same backbone (MNetV3-Large) baseline future study.