Clinton Fookes

Deep learning models can accurately predict urban land surface temperature, but predictive performance does not ensure reliable estimates of interventions. This study evaluates the counterfactual reliability of TITAN-3D CR, a nine-channel deep learning model incorporating counterfactual consistency regularisation, using 668 held-out tiles across four Adelaide regions. Three findings expose important inferential limitations. First, conditioning on a post-treatment canopy-related covariate reduces the observed tree-reference temperature differential from 4.81°C to 2.3°C, whereas defensible pre-treatment adjustment reduces it by only 7.6%. Second, model-implied warming following tree removal varies from 3.25-5.39°C across defensible counterfactual constructions, with construction uncertainty approximately 14 times greater than sampling uncertainty. Third, despite 99.4% matched-scene constraint satisfaction, isolated synthetic interventions remain strongly seed dependent. These results show that counterfactual urban thermal modelling requires systematic evaluation of intervention construction, model stability, and uncertainty. Such evaluation is essential for developing trustworthy modelling tools to support urban heat adaptation decisions.

Neural Radiance Fields (NeRFs) have remodeled 3D scene representation since release. NeRFs can effectively reconstruct complex 3D scenes from 2D images, advancing different fields and applications such as scene understanding, 3D content generation, and robotics. Despite significant research progress, a thorough review of recent innovations, applications, and challenges is lacking. This survey compiles key theoretical advancements and alternative scene representations and investigates emerging challenges. It further explores applications on reconstruction, highlights NeRFs’ impact on computer vision and robotics, and reviews essential datasets and toolkits. By identifying gaps in the literature, this survey discusses open challenges and offers directions for future research.

Person re-identification (Re-ID) across visible and infrared modalities is crucial for 24-hour surveillance systems, but existing datasets primarily focus on ground-level perspectives. While ground-based IR systems offer nighttime capabilities, they suffer from occlusions, limited coverage, and vulnerability to obstructions—problems that aerial perspectives uniquely solve. To address these limitations, we introduce AG-VPReID.VIR, the first aerial-ground cross-modality video-based person Re-ID dataset. This dataset captures 1,837 identities across 4,861 tracklets (124,855 frames) using both UAV-mounted and fixed CCTV cameras in RGB and infrared modalities. AG-VPReID.VIR presents unique challenges including cross-viewpoint variations, modality discrepancies, and temporal dynamics. We propose TCC-VPReID, a novel three-stream architecture designed to address the joint challenges of cross-platform and cross-modality person Re-ID. We enhance this with ODE-Fusion, a neural ODE framework for fusing paired RGB-IR captures from the same platform, facilitating effective cross-modal feature fusion through continuous state evolution. Our approach bridges the domain gaps through style-robust feature learning, memory-based cross-view adaptation, intermediary-guided temporal modeling, and ODE-based fusion. Experiments show that AG-VPReID.VIR presents distinctive challenges compared to existing datasets, with our enhanced TCC-VPReID framework achieving significant performance gains across multiple evaluation protocols. Dataset and code will be made available.