arXiv Computer Vision

X Marks the Spot: Diffusion for Image Geolocalization with Location Representations

The paper introduces a new diffusion-based approach for image geolocalization that learns in the pretrained representation space of location encoders rather than directly predicting coordinates. A Location Autoencoder (LAE) maps these embeddings back to coordinates with kilometer-level accuracy, enabling a latent diffusion model (DIG) to transport noise to location embeddings conditioned on image features. DIG outperforms previous coordinate-based generative models and sets a new state of the art on three major benchmarks, becoming the first generative model to surpass retrieval-based methods and improving retrieve‑and‑rerank systems when used as a prior.

arXiv Computer Vision
Aug 27

GTPred: Benchmarking MLLMs for Interpretable Geo-localization and Time-of-capture Prediction

GTPred is a new benchmark for geo‑temporal prediction that evaluates multi‑modal large language models (MLLMs) on 370 images taken across 120 years worldwide. It assesses predictions by matching both the year and a hierarchical location sequence, and includes annotated reasoning chains to test intermediate reasoning. Experiments on 15 MLLMs show that while visual perception is strong, models still lack world knowledge and geo‑temporal reasoning, and that adding temporal data improves location inference.

By Jinnao Li, Tingzhu Chen, Changbo Wang
arXiv Computer Vision
Nov 6

GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation

GeoCore-9B is a 9‑billion‑parameter generative foundation model trained from scratch on Earth observation data, using a Flow Matching‑based Diffusion Transformer that conditions on text and continuous geospatial metadata. It introduces a Geospatial Semantic Alignment loss to guide training with a frozen specialist teacher, enabling the model to learn structural Earth surface priors without inference overhead. Evaluations show GeoCore-9B achieves state‑of‑the‑art visual fidelity and geographic structural accuracy, and can be adapted to practical EO tasks such as cloud removal and SAR‑to‑optical translation.

By Jeonghyeok Do, Munchurl Kim
arXiv Machine Learning
Sep 29

Do Location Encoders Capture Spatial Effects? A GeoShapley Benchmark Across Scales

Location encoders transform geographic coordinates into high‑dimensional embeddings for machine learning, yet it is unclear how well these embeddings capture interpretable spatial effects. This study benchmarks GeoShapley—a game‑theoretic explainer treating all location features as a single joint player—against eleven TorchSpatial encoders on a synthetic process with known coefficients, across grid, county, and global scales, with and without raw coordinates and under different training regimes. The results show that primary coefficient recovery is consistently high across encoders, while secondary coefficient recovery varies more with scale, especially at the global level, and raw‑coordinate baselines remain competitive throughout.

By Daniel Kiv, Shaowen Wang
arXiv Computer Vision
Oct 15

GeoThinker-$F$: Endogenous Predictive World Representations for Fast Spatial Intelligence

arXiv:2610.20592v1 Announce Type: new Abstract: Existing geometry-aware Multimodal Large Language Models (MLLMs) increasingly rely on external 3D encoders to provide structural priors, but executing...

By Haoyuan Li, Qihang Cao, Huanyuan Zhou, Xin Hu, Sijia Chen, Zihan Guo, Kun Xiang, Yunshuang Nie, Tao Tang, Xinlong Zheng, Jia-Wang Bian, Xianming Liu, Xiaodan Liang