Multiexposure image fusion (MEF) provides a simple, effective, and low-cost solution to fill the gap between the high dynamic range of natural scenes and the low dynamic range of commonly used imaging sensors. In recent years, deep learning has made remarkable advances in MEF. However, it remains relatively challenging to make a fused image retain more detailed information from source images while making the brightness of the fused image more uniform. In this article, the multiscale interaction network multiscale interaction network (MIN)-MEF is proposed to fuse two extremely exposed images. In MIN-MEF, a dual-branch interaction extractor (DBIE) based on a dual attention block (DAB) and feature interaction block (FIB) is designed to fully extract the features of under-/over-exposed images. To balance global semantic information and local detail features during the fusion process, we perform feature extraction and integration at different scales by Gaussian pyramid decomposition and the proposed reconstruction module. Moreover, to ensure that the fusion results have more uniform luminance and richer details, we design an improved pixel loss by defining a luminance modulation coefficient and a gradient decision coefficient to modulate the luminance and detail information of the source images, respectively. Extensive experiments on two publicly available datasets show that the proposed MIN-MEF outperforms state-of-the-art methods in subjective and objective evaluations.