# Adaptive Cross-Modal Fusion with Sparse Attention for Pedestrian Crossing Intention Prediction

arXiv: https://arxiv.org/abs/2607.12293 · HTML: https://mlsanity.com/paper/2607.12293/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 17 of its 40 references that are in the map.
- First posted on arXiv 14 Jul 2026.

## Builds on (most cited first, 17 of 17)

- [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://mlsanity.com/paper/2312.00752/) (2023)
- [Swin Transformer V2: Scaling Up Capacity and Resolution](https://mlsanity.com/paper/2111.09883/) (2021)
- [SimMIM: A Simple Framework for Masked Image Modeling](https://mlsanity.com/paper/2111.09886/) (2021)
- [Interpretable End-to-end Urban Autonomous Driving with Latent Deep Reinforcement Learning](https://mlsanity.com/paper/2001.08726/) (2020)
- [Autonomous Vehicles that Interact with Pedestrians: A Survey of Theory and Practice](https://mlsanity.com/paper/1805.11773/) (2018)
- [Pedestrian Action Anticipation using Contextual Feature Fusion in Stacked RNNs](https://mlsanity.com/paper/2005.06582/) (2020)
- [Predicting Pedestrian Crossing Intention with Feature Fusion and Spatio-Temporal Attention](https://mlsanity.com/paper/2104.05485/) (2021)
- [Agreeing to Cross: How Drivers and Pedestrians Communicate](https://mlsanity.com/paper/1702.03555/) (2017)
- [Interaction-aware Decision-making for Automated Vehicles using Social Value Orientation](https://mlsanity.com/paper/2207.05853/) (2022)
- [Intention Recognition of Pedestrians and Cyclists by 2D Pose Estimation](https://mlsanity.com/paper/1910.03858/) (2019)
- [FuSSI-Net: Fusion of Spatio-temporal Skeletons for Intention Prediction Network](https://mlsanity.com/paper/2005.07796/) (2020)
- [IntFormer: Predicting pedestrian intention with the aid of the Transformer architecture](https://mlsanity.com/paper/2105.08647/) (2021)
- [TrouSPI-Net: Spatio-temporal attention on parallel atrous convolutions and U-GRUs for skeletal pedestrian crossing prediction](https://mlsanity.com/paper/2109.00953/) (2021)
- [Deep Virtual-to-Real Distillation for Pedestrian Crossing Prediction](https://mlsanity.com/paper/2211.00856/) (2022)
- [TAMFormer: Multi-Modal Transformer with Learned Attention Mask for Early Intent Prediction](https://mlsanity.com/paper/2210.14714/) (2022)
- [Multi-Context Fusion Transformer for Pedestrian Crossing Intention Prediction in Urban Environments](https://mlsanity.com/paper/2511.20011/) (2025)
- [ACIT: Attention-Guided Cross-Modal Interaction Transformer for Pedestrian Crossing Intention Prediction](https://mlsanity.com/paper/2511.20020/) (2025)

## Abstract (arXiv)

Predicting pedestrian crossing intention is a safety-critical task for autonomous driving, yet existing approaches often rely on single-modal inputs or dense multimodal fusion strategies that inadequately capture complementary visual and kinematic information while introducing redundant inter-modal interactions. We propose ADAPT (Adaptive Domain-Aware Pedestrian Crossing Transformer), a multimodal framework that jointly models local and global visual context together with temporal motion dynamics for accurate pedestrian crossing intention prediction. ADAPT processes four spatially aligned visual modalities, including RGB images, local depth maps, global semantic maps, and global depth maps, together with ego-vehicle speed, pedestrian bounding boxes, and skeleton pose information through five specialized modules: a weight-shared Swin Transformer V2 backbone for visual feature extraction, a Cross-Modality Guided Attention module for hierarchical visual fusion, a Mamba-based Motion Feature Encoding module for efficient temporal modeling, a Sparse Cross-Modal Attention module that selectively preserves the most informative inter-modal interactions, and a Vision Transformer-based Temporal Feature Fusion module for sequence-level prediction. Extensive experiments on the JAAD and PIE benchmark datasets demonstrate that ADAPT consistently outperforms existing state-of-the-art methods while maintaining low computational complexity. On JAAD, the proposed method achieves an AUC of 0.73 on JAADbeh and 0.85 on JAADall, while on PIE it achieves an accuracy of 0.92 and an AUC of 0.90. Furthermore, ADAPT performs inference in only 17.23 ms per sample, offering an effective balance between predictive accuracy and real-time deployment efficiency for intelligent transportation and autonomous driving applications.