Research map / Vision-language and multimodal / Cross-modal learning
MapFormer: Boosting Change Detection by Using Pre-change Information
Maximilian Bernhard, Niklas Strauß, Matthias Schubert
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 |
|---|---|---|---|
| add pre-change semantic map as input (concatenation baseline) | Concatenation change detector · DynamicEarthNet | binary change IoU: 11.6 -> 20.1 | Improved |
| add pre-change semantic map as input (concatenation baseline) | Concatenation change detector · HRSCD | binary change IoU: 25.9 -> 45.1 | Improved |
| use MapFormer fusion module instead of plain concatenation | Conditional BCD · DynamicEarthNet | binary change IoU: 20.1 -> 23.5 | Improved |
| use MapFormer fusion instead of DynamicMLP | Conditional BCD · HRSCD | binary change IoU: 45.7 -> 47.7 | Improved |
| add supervised cross-modal contrastive loss (ablation) | MapFormer K=10 · DynamicEarthNet | binary change IoU: 20.9 -> 23.5 | Improved |
| add stop-gradient on map features in contrastive loss (ablation) | MapFormer K=10 · DynamicEarthNet | binary change IoU: 22.5 -> 23.5 | Improved |
| do not add linear projection on map features (ablation) | MapFormer K=10 · HRSCD | binary change IoU: 45.9 -> 47.7 | Improved |
| use full-class semantic map instead of merged high-level classes (ablation) | MapFormer K=10 · DynamicEarthNet | binary change IoU: 17.6 -> 23.5 | Improved |
| use full-resolution semantic map instead of 32x downsampled (ablation) | MapFormer K=10 · HRSCD | binary change IoU: 43.2 -> 47.7 | Improved |
| vary hyperparameter K (5, 10, 15) (ablation) | MapFormer · HRSCD | binary change IoU: 48.0 / 47.7 / 47.6 | No clear effect |
| use predicted maps instead of true pre-change map (w/o m1) | MapFormer K=10 · DynamicEarthNet (bi-temporal) | binary change IoU: 11.8 -> 16.7 | Improved |
| drop pre-change image (cross-modal) and keep pre-change map (ablation) | MapFormer · DynamicEarthNet | binary change IoU: 23.5 -> 19.0 | Hurt |
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)
- RSRCC: A Remote Sensing Regional Change Comprehension Benchmark Constructed via Retrieval-Augmented Best-of-𝑁 Ranking (NeurIPS 2026 · similarity 0.88)
- 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)
- UniChange: Unifying Change Detection with Multimodal Large Language Model (CVPR 2026 · similarity 0.87)
- The STVchrono Dataset: Towards Continuous Change Recognition in Time (CVPR 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.
