PECR:Parameter-Efficient Transfer Learning with Cross-Modal Representation Learning for Remote Sensing Visual Question Answering

Remote sensing (RS) visual question answering (VQA) aims to provide accurate answers to questions related to RS images. Transformer-based models have gradually become popular to solve RS VQA tasks. Due to the ever-growing model size, full-parameter training of the model becomes prohibitively costly. Moreover, most current RS VQA methods primarily focus on improving unimodal image encoding, paying little attention to cross-modal interactions between visual and textual features. In this paper, we propose the Parameter-Efficient transfer learning with Cross-modal Representation learning model (PECR) for RS VQA tasks. Specifically, we introduce adapter-based parameter-efficient transfer learning techniques into the visual encoder and initialize them with pre-trained weights on large-scale RS images. Furthermore, we utilize a cross attention mechanism in the cross-modal fusion module to merge context representations of images and text, facilitating cross-modal representation learning. Experimental results demonstrate that our approach outperforms previous state-of-the-art methods on both RSVQA-LR and RSVQA-HR datasets. Additionally, we also validate that employing the adapter strategy for local parameter training can yield performance results comparable to full parameter training, significantly reducing the model training cost.