Accurate segmentation of infant brain images from magnetic resonance imaging (MRI) scans is crucial for studying brain development. Existing deep learning methods often rely on encoder-decoder structures with local operators, limiting their ability to efficiently capture long-range information. Moreover, these models struggle to integrate diverse tissue properties from different MRI sequences, leading to computational and memory challenges during inference. To address these limitations, we propose a novel model, 3D-DenseUNet, which incorporates adaptable global aggregation blocks to mitigate spatial information loss during down-sampling. We enhance the model with a self-attention module that integrates feature maps across spatial and channel dimensions, improving representation potential and discrimination ability. Furthermore, we introduce a novel learning scheme, termed "two independent teachers," which leverages model weights instead of label predictions. In this scheme, each teacher model is trained on a specific MRI sequence (T1 and T2) to capture diverse tissue properties. A fusion model is then employed to enhance test accuracy while reducing computational overhead. Empirical evaluations on two distinct datasets demonstrate the effectiveness of our approach. Our source code is publicly available at https://github.com/AfifaKhaled/Two-Independent-Teachers-are-Better-Role-Model .