GeoTransformer: Enhancing Urban Forecasting with Geospatial Attention Mechanisms

Recent advancements have focused on encoding urban spatial information into high-dimensional spaces, with notable efforts dedicated to integrating sociodemographic data and satellite imagery. These efforts have established foundational models in this field. However, the effective utilization of these spatial representations for urban forecasting applications remains under-explored. To address this gap, we introduce GeoTransformer, a novel structure that synergizes the Transformer architecture with geospatial statistics prior. GeoTransformer employs an innovative geospatial attention mechanism to incorporate extensive urban information and spatial dependencies into a unified predictive model. Specifically, we compute geospatial weighted attention scores between the target region and surrounding regions and leverage the integrated urban information for predictions. Extensive experiments on GDP and ride-share demand prediction tasks demonstrate that GeoTransformer significantly outperforms existing baseline models, showcasing its potential to enhance urban forecasting tasks.

DivCon: Divide and Conquer for Progressive Text-to-Image Generation

Diffusion-driven text-to-image (T2I) generation has achieved remarkable advancements. To further improve T2I models' capability in numerical and spatial reasoning, the layout is employed as an intermedium to bridge large language models and layout-based diffusion models. However, these methods still struggle with generating images from textural prompts with multiple objects and complicated spatial relationships. To tackle this challenge, we introduce a divide-and-conquer approach which decouples the T2I generation task into simple subtasks. Our approach divides the layout prediction stage into numerical \& spatial reasoning and bounding box prediction. Then, the layout-to-image generation stage is conducted in an iterative manner to reconstruct objects from easy ones to difficult ones. We conduct experiments on the HRS and NSR-1K benchmarks and our approach outperforms previous state-of-the-art models with notable margins. In addition, visual results demonstrate that our approach significantly improves the controllability and consistency in generating multiple objects from complex textural prompts.