Reproduce “Open-Set Face Recognition with Maximal Entropy and Objectosphere Loss” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rafael Henrique Vareto, Yu Linghu, Terrance E. Boult, William Robson Schwartz, Manuel Günther
A new method enhances open-set face recognition by integrating a pre-trained deep neural network with an adapter network and a novel Maximal Entropy Loss function. This approach improves the discrimination between known and unknown identities, demonstrating improved performance on challenging datasets such as IJB-C and UCCS.
Key insights
- The proposed Maximal Entropy Loss (MEL) consistently improves open-set recognition performance, particularly with appropriately selected negative samples.
- The inclusion and strategic selection of negative samples from a similar distribution are crucial for enhancing open-set recognition and rejection capabilities.
- An adapter network provides an efficient means to adapt pre-trained deep neural networks for open-set face recognition without requiring full backbone retraining, sometimes even outperforming full finetuning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.