Technical Approach for the EMI Challenge in the 8th Affective Behavior Analysis in-the-Wild Competition

TL;DR

The paper addresses the challenge of estimating Emotional Mimicry Intensity (EMI) from multifaceted, time-varying signals. It introduces a dual-stage framework that first performs vision-text and audio-text contrastive alignment via CLIP-based networks, then builds a tri-modal temporal model with Temporal Convolutional Networks for visual features, BiLSTM for audio, and a quality-guided fusion mechanism to adaptively weight modalities. Key contributions include the Stage I and Stage II cross-modal alignment, temporal feature enhancement, and a differentiable, noise-robust fusion strategy that yields strong performance on the Hume-Vidmimic2 dataset, with $\rho$ values of $0.51$ on validation and $0.68$ on the test set, securing runner-up in the EMI Track of the 8th ABAW Competition. This approach demonstrates a practical pathway for fine-grained EMI estimation in unconstrained environments, with implications for robust affective computing in human-computer interaction.

Abstract

Emotional Mimicry Intensity (EMI) estimation plays a pivotal role in understanding human social behavior and advancing human-computer interaction. The core challenges lie in dynamic correlation modeling and robust fusion of multimodal temporal signals. To address the limitations of existing methods--insufficient exploitation of cross-modal synergies, sensitivity to noise, and constrained fine-grained alignment capabilities--this paper proposes a dual-stage cross-modal alignment framework. Stage 1 develops vision-text and audio-text contrastive learning networks based on a CLIP architecture, achieving preliminary feature-space alignment through modality-decoupled pre-training. Stage 2 introduces a temporal-aware dynamic fusion module integrating Temporal Convolutional Networks (TCN) and gated bidirectional LSTM to capture macro-evolution patterns of facial expressions and local dynamics of acoustic features, respectively. A novel quality-guided fusion strategy further enables differentiable weight allocation for modality compensation under occlusion and noise. Experiments on the Hume-Vidmimic2 dataset demonstrate superior performance with an average Pearson correlation coefficient of 0.51 across six emotion dimensions on the validate set. Remarkably, our method achieved 0.68 on the test set, securing runner-up in the EMI Challenge Track of the 8th ABAW (Affective Behavior Analysis in the Wild) Competition, offering a novel pathway for fine-grained emotion analysis in open environments.

Figures (1)

• Figure 1: The overview of our proposed method. We first utilize the three encoders with frozen parameters pre-trained in the first stage, the audio, image, and text are then fed into these three encoders to obtain $F_A$, $F_I$, $F_T$. After that, $F_I$ and $F_A$ will be strengthened by temporal features for the Temporal Feature Enhancer (TFE), the temporally enhanced features are then processed by the Quality-Aware Module (QAM), which outputs weights for each feature. These weights are subsequently applied to the corresponding features, and the weighted features from the three modalities (audio, visual, text) are concatenated and fed into the Transformer Decoder module followed by the Classifier module to generate the final prediction.