Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study investigates the challenges of Visual Question Answering (VQA) in the Remote Sensing (RS) domain and evaluates the effectiveness of a Parameter Efficient Fine Tuning (PEFT) strategy called RS Adapter across three Vision Language Model architectures: Dual Encoder CLIP, Encoder Decoder BLIP, and Hybrid FLAVA. By implementing a unified architectural surgery pipeline that integrates lightweight bottleneck adapters into frozen backbones, the authors achieve rapid adaptation with minimal trainable parameters. The results reveal that the Hybrid FLAVA model outperforms others in multimodal reasoning and retrieval, setting a new benchmark for efficient VQA applications in disaster assessment and urban monitoring.
Hybrid FLAVA achieves superior multimodal reasoning in Remote Sensing VQA with less than 5% of trainable parameters, redefining efficiency in the field.
Visual Question Answering (VQA) in the Remote Sensing (RS) domain presents unique challenges due to the high resolution, multi scale object distribution, and semantic complexity of aerial imagery. While general domain Foundation Models have achieved remarkable success, their direct application to RSVQA is hindered by massive domain shifts and the computationally prohibitive nature of full fine tuning. This study presents a comparative analysis of RS Adapter, a Parameter Efficient Fine Tuning (PEFT) strategy, applied across three distinct Vision Language Model (VLM) architectures: the Dual Encoder CLIP, the Encoder Decoder BLIP, and the Hybrid FLAVA. We introduce a unified architectural surgery pipeline that injects lightweight bottleneck adapters into the attention and MLP layers of frozen backbones, enabling rapid adaptation with less than 5 percent of trainable parameters. Experimental results on the high resolution RSVQA x dataset demonstrate that while all adapted models achieve convergence, the Hybrid FLAVA architecture offers a superior balance of multimodal reasoning and retrieval capabilities compared to its unimodal counterparts. Our findings establish a new baseline for resource efficient VQA in disaster assessment and urban monitoring.