MoodPupilar introduces a novel method for mood evaluation using pupillary response captured by a smartphone's front- facing camera during daily use. Over a four-week period, data was gathered from 25 participants to develop models capable of predicting daily mood averages. Utilizing the GLOBEM behavior modeling platform, we benchmarked the utility of pupillary response as a predictor for mood. Our proposed model demonstrated a Matthew's Correlation Coefficient (M CC) score of 0.15 for Valence and 0.12 for Arousal, which is on par with or exceeds those achieved by existing behavioral modeling algorithms supported by GLOBEM. This capability to accurately predict mood trends underscores the effectiveness of pupillary response data in providing crucial insights for timely mental health interventions and resource allocation. The outcomes are encouraging, demonstrating the potential of real-time and pre-dictive mood analysis to support mental health interventions.

Traditional semi-supervised learning tasks assume that both labeled and unlabeled data follow the same class distribution, but the realistic open-world scenarios are of more complexity with unknown novel classes mixed in the unlabeled set. Therefore, it is of great challenge to not only recognize samples from known classes but also discover the unknown number of novel classes within the unlabeled data. In this paper, we introduce a new open-world semi-supervised novel class discovery approach named OpenNCD, a progressive bi-level contrastive learning method over multiple prototypes. The proposed method is composed of two reciprocally enhanced parts. First, a bi-level contrastive learning method is introduced, which maintains the pair-wise similarity of the prototypes and the prototype group levels for better representation learning. Then, a reliable prototype similarity metric is proposed based on the common representing instances. Prototypes with high similarities will be grouped progressively for known class recognition and novel class discovery. Extensive experiments on three image datasets are conducted and the results show the effectiveness of the proposed method in open-world scenarios, especially with scarce known classes and labels.