<p>Mask-occluded face recognition (MOFR) methods aim to activate highly relevant spatial locations and feature channels for learning robust identity features. However, most existing methods model spatial-wise information without learning interaction among features at different scales that contain information with different granularity, leading to imprecise localization of informative non-masked spatial positions. Furthermore, the existing methods overlook the interaction among different feature channels that gather information from the same spatial position but react differently to occlusions. Based on the observation that the image scale is not exclusive to the spatial attention, and the channel attention can also possess multiple scales other than the global, the multi-scale context information can also be aggregated along the channel dimension. Based on the abovementioned observations, we propose a two-stage interactive feature learning module (IFLM), consisting of two submodules, namely the hybrid channel attention module (HCAM) and the cascaded spatial and channel feature interaction module (SCFIM). In the first stage, our HCAM calibrates the channel-wise feature responses by aggregating both local and global feature context, addressing the issue of scale variation from a channel context. The resultant channel-wise attentive feature map then acts as input to SCFIM, which builds connections across globally and locally distributed spatial and channel information in a cascaded fashion to learn corresponding relationship matrices via a self-attention mechanism, resulting in highlighting effective spatial locations and feature channels for MOFR. The effectiveness of IFLM is validated through a comprehensive evaluation on various publicly available mask-occluded face datasets. The source code of our proposed method is available at <a href="https://github.com/SaadShakeel414/IFLM.">https://github.com/SaadShakeel414/IFLM.</a></p>