LibEMER: A novel benchmark and algorithms library for EEG-based Multimodal Emotion Recognition

TL;DR

LibEMER tackles the lack of open-source tools and standardized benchmarks in EEG-based multimodal emotion recognition by delivering a PyTorch-based, end-to-end library with reproducible implementations and unified preprocessing, data splitting, and evaluation protocols. It benchmarks 10 representative models (from DNN, CNN, RNN, Transformer, and GNN families) across three datasets (SEED, SEEDV, DEAP) and two tasks (subject-dependent and subject-independent), revealing that task design and cross-modal fusion strategies often outperform architectural choices. Reproducibility experiments show most reproduced results stay within a $10\%$ gap of reported figures, highlighting both progress and the impact of reporting and platform variability. The analysis underscores the importance of standardized benchmarks and larger, more diverse datasets to overcome data sparsity and variability, guiding future EMER research toward more reliable and generalizable models. LibEMER thus accelerates standardization and transparency in EMER and offers a practical path for developing and evaluating new multimodal emotion recognition algorithms.

Abstract

EEG-based multimodal emotion recognition(EMER) has gained significant attention and witnessed notable advancements, the inherent complexity of human neural systems has motivated substantial efforts toward multimodal approaches. However, this field currently suffers from three critical limitations: (i) the absence of open-source implementations. (ii) the lack of standardized and transparent benchmarks for fair performance analysis. (iii) in-depth discussion regarding main challenges and promising research directions is a notable scarcity. To address these challenges, we introduce LibEMER, a unified evaluation framework that provides fully reproducible PyTorch implementations of curated deep learning methods alongside standardized protocols for data preprocessing, model realization, and experimental setups. This framework enables unbiased performance assessment on three widely-used public datasets across two learning tasks. The open-source library is publicly accessible at: https://anonymous.4open.science/r/2025ULUIUBUEUMUEUR485384

Figures (2)

• Figure 1: Two primary tasks for emotion recognition
• Figure 2: Scores of model performance on two tasks based on Accuracy and F1-score