TRUFM: a Transformer-Guided Framework for Fine-Grained Urban Flow Inference

Reconstructing the fine-grained urban flow from the coarse-grained counterpart is an essential component in intelligent transportation systems, as it can provide accurate traffic flow information under a reduced number of sensors. However, current models based on Convolutional Neural Networks (CNNs) mainly focus on the local pixel correlations and ignore the long-range dependencies. To this end, we propose a TRansformer-guided Urban Flow Magnifier (TRUFM) that incorporates the transformer module in the traffic flow analysis system, which naturally enjoys the advantage of modeling the global-scale correlations. By utilizing this superiority, our framework facilitates the joint inference of the flow distribution across the entire map and hence estimates more precise fine-grained traffic flow. Experimental results demonstrate the effectiveness of our TRUFM, which exceeds the current state-of-the-art methods on various datasets.