CMFA_Net: A cross-modal feature aggregation network for infrared-visible image fusion

•The channel-spatial-wise attention convolutional (CSAC) layer is put forward and integrated into the feature extractor to extract effective features and focus on the affinity areas of dual modalities.•A feature aggregation strategy based on the attention mechanism and l1-norm is proposed for fusing the depth features appropriately.•We demonstrate that substituting batch normalization (BN) with group normalization (GN) layer can accelerate the training of the network and meanwhile prevent overfitting.•A specific loss function is imposed for training the model, which is composed of SSIM-p and CGV to enable the fused image equilibrate high-quality and abundent background details.