Researchers have introduced CrossTimeEdit, a framework for generating historical street views, alongside VIGOR-his, a decade-spanning cross-view dataset, according to a paper published on Hugging Face. The work addresses a persistent problem: historical street-view imagery documents urban change, but uneven coverage leaves substantial gaps in the historical record.
The VIGOR-his dataset contains 43,653 location-level quadruplets spanning 11 cities on three continents. Its automated pipeline performs spatial pairing, consistency screening, change classification, and the generation and validation of satellite-based change descriptions and local editing instructions.
Rather than generating past street views from scratch, CrossTimeEdit reformulates the task as editing. It uses recent street views to constrain viewpoint and unchanged appearance, while temporal satellite differences serve as evidence of change. The model starts from FLUX.2 [Klein] 4B and is trained through supervised fine-tuning followed by online reinforcement learning.
The authors design three street-view editing criteria — Instruction Alignment (IA), Background Preservation (BP), and Quality and Physical Plausibility (QP) — serving as both RL reward dimensions and evaluation metrics. The multi-reward objective is optimized using Within Group Relative Policy Optimization for flow-matching models (Flow-GRPO) with Group reward-Decoupled Normalization Policy Optimization (GDPO), which normalizes each reward dimension before aggregation.
According to the paper, CrossTimeEdit improves overall performance across the three editing criteria by 17.12% over the pretrained baseline and outperforms cross-view generation models in scene consistency, visual realism, and perceptual quality. The implementation code, dataset, and model weights are publicly available.