Data augmentation for deep visual recognition using superpixel based pairwise image fusion

Data augmentation is an important paradigm for boosting the generalization capability of deep learning in image classification tasks. Image augmentation using cut-and-paste strategies has shown very good performance improvement for deep learning. However, these existing methods often overlook the image's discriminative local context and rely on ad hoc regions consisting of square or rectangular local regions, leading to the loss of complete semantic object parts. In this work, we attempt to overcome these limitations and propose a superpixel-wise local-context-aware efficient image fusion approach for data augmentation. Our approach requires only one forward propagation using a superpixel attention-based label fusion with less computational complexity. The model is trained using a combination of a global classification of the fused (augmented) image loss, a superpixel-wise weighted local classification loss, and a superpixel-based weighted contrastive learning loss. The last two losses are based on the superpixel-aware attentive embeddings. Thus, the resulting deep encoder can learn both local and global features of the images while capturing object-part local context and information. Experiments on diverse benchmark image datasets indicate that the proposed method out-performs many region-based augmentation methods for visual recognition. We have demonstrated its effectiveness not only on CNN models but also on transformer models. The codes are accessible at https://github.com/DanielaPlusPlus/SAFuse.