Native multimodal Qwen models outperform structured vision-language pipelines on the CDVQA benchmark for change VQA in remote sensing, with performance not scaling monotonically with model size.
Kernel-adaptive change detection network in remote sensing imagery
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Revisiting Change VQA in Remote Sensing with Structured and Native Multimodal Qwen Models
Native multimodal Qwen models outperform structured vision-language pipelines on the CDVQA benchmark for change VQA in remote sensing, with performance not scaling monotonically with model size.