Research map / Time series / Foundation models for time series
Changen2: Multi-Temporal Remote Sensing Generative Change Foundation Model
Zhuo Zheng, Stefano Ermon, Dongjun Kim, Liangpei Zhang, Yanfei Zhong
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Changen with Changen2 diffusion-transformer generator | ChangeStar R-18 zero-shot · LEVIR-CD all | F1: 62.8 -> 83.7 | Improved |
| replace Changen with Changen2 generator | ChangeStar R-18 zero-shot · WHU-CD all | F1: 26.6 -> 76.3 | Improved |
| use pretrained Stable Diffusion ControlNet instead of DiT trained from scratch | GPCM generators · xView2 hold | FID: 32.33 -> 101.45 | Hurt |
| use Changen2 instead of Changen for image quality | Changen · xView2 hold | FID: 34.74 -> 32.44 | Improved |
| scale change detector backbone from R-18 to ViT-B under zero-shot | Changen2 pretraining · LEVIR-CD | F1: 83.7 -> 88.7 | Improved |
| pretrain on Changen2 synthetic data instead of xView2 segmentation data (ablation) | ChangeStar R-18 · LEVIR-CD test | F1: 90.6 -> 91.3 | Improved |
| pretrain on Changen2 instead of Changen | ChangeStar MiT-B1 · LEVIR-CD test | F1: 91.5 -> 91.9 | Improved |
| pretrain on Changen2 instead of SA-1B | ChangeStar ViT-B · S2Looking test | F1: 68.1 -> 68.5 | Improved |
| use smaller pre-event guidance ratio lambda (more temporal diversity) (ablation) | Changen2 data · zero-shot LEVIR-CD, S2Looking, others | F1 | Improved |
| pretrain on Changen2 instead of OpenEarthMap (ablation) | ChangeStar ViT-B · SECOND test | SeK37: 21.2 -> 23.0 | Improved |
| pretrain on Changen2 instead of OpenEarthMap (ablation) | ChangeStar EfficientNet-B0 · SECOND test | mIoU30: 27.5 -> 26.2 | Mixed |
| zero-shot Changen2 ChangeStar versus AnyChange | ChangeStar ViT-B · SECOND BCD | F1: 44.6 -> 61.0 | Improved |
Closest papers by meaning
- The Change You Want To Detect: Semantic Change Detection In Earth Observation With Hybrid Data Generationf (CVPR 2025 · similarity 0.91)
- FM-ChangeNet: Learning Change through Pathwise Feature Transport (NeurIPS 2026 · similarity 0.88)
- Feature Spectrum Learning for Remote Sensing Change Detection (CVPR 2025 · similarity 0.88)
- RSRCC: A Remote Sensing Regional Change Comprehension Benchmark Constructed via Retrieval-Augmented Best-of-𝑁 Ranking (NeurIPS 2026 · similarity 0.87)
- MMM-RS: A Multi-modal, Multi-GSD, Multi-scene Remote Sensing Dataset and Benchmark for Text-to-Image Generation (NeurIPS 2024 · similarity 0.87)
- DiffusionSat: A Generative Foundation Model for Satellite Imagery (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
