Open-Set Face Recognition with Maximal Entropy and Objectosphere Loss

Rafael Henrique Vareto,Yu Linghu,Terrance E. Boult,William Robson Schwartz,Manuel Günther

2023-11-01

Abstract:Open-set face recognition characterizes a scenario where unknown individuals, unseen during the training and enrollment stages, appear on operation time. This work concentrates on watchlists, an open-set task that is expected to operate at a low False Positive Identification Rate and generally includes only a few enrollment samples per identity. We introduce a compact adapter network that benefits from additional negative face images when combined with distinct cost functions, such as Objectosphere Loss (OS) and the proposed Maximal Entropy Loss (MEL). MEL modifies the traditional Cross-Entropy loss in favor of increasing the entropy for negative samples and attaches a penalty to known target classes in pursuance of gallery specialization. The proposed approach adopts pre-trained deep neural networks (DNNs) for face recognition as feature extractors. Then, the adapter network takes deep feature representations and acts as a substitute for the output layer of the pre-trained DNN in exchange for an agile domain adaptation. Promising results have been achieved following open-set protocols for three different datasets: LFW, IJB-C, and UCCS as well as state-of-the-art performance when supplementary negative data is properly selected to fine-tune the adapter network.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper primarily aims to address specific issues in Open-set Face Recognition, particularly focusing on the low false positive recognition rate in "Watchlists" scenarios. In this scenario, individuals not seen during the training phase appear in real-world operations, and the registered samples for each identity are usually very few.

Key contributions of the paper include:
1. **Proposing a compact adapter network**: This network can receive facial feature representations from a pre-trained deep neural network and act as a substitute for the output layer to achieve agile domain adaptation.
2. **Introducing Maximal Entropy Loss (MEL)**: This is a new loss function designed to increase the entropy of negative samples and add penalties to known target classes, thereby pushing the network to specialize in the gallery set.
3. **Investigating Objectosphere Loss**: Analyzing how this loss modifies the feature vector norms of training and testing facial samples.
4. **Detailed evaluation of different loss functions**: Conducting exhaustive open-set analysis on datasets containing thousands of identities but few samples per class.
5. **Experimental validation of the method's effectiveness**: Experiments were conducted on well-known open-set face recognition datasets by combining different deep feature extractors.

The paper uses three different datasets for evaluation: LFW, IJB-C, and UCCS. These datasets have thousands of identities but very few samples per identity, making them ideal for studying open-set face recognition problems.

The paper also provides a detailed description of the proposed pipeline structure, including feature extraction, adaptation modules, and classification subsystems, and explores the effects of different loss functions. Additionally, it offers detailed experimental results demonstrating the performance of the proposed method on different datasets.