Research map / Time series / Foundation models for time series
ViTs for SITS: Vision Transformers for Satellite Image Time Series
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use temporal-then-spatial rather than spatial-then-temporal factorization (ablation) | TSViT · Germany | mIoU: 48.8 -> 78.5; +29.7% | Helped |
| Add K class-specific cls tokens instead of one (ablation) | TSViT · Germany | mIoU: 78.5 -> 83.6; +5.1% | Helped |
| Use acquisition-date lookup rather than static position encodings (ablation) | TSViT · Germany | mIoU: 80.8 -> 83.6; -2.8% performance drop with fixed encodings | Helped |
| Block cls-token interactions in the spatial encoder (ablation) | TSViT · Germany | mIoU: 81.5 -> 83.6; -2.1% mIoU when interactions are allowed | Helped |
| Use 2 × 2 rather than 3 × 3 patches (ablation) | TSViT · Germany | mIoU: 83.6 -> 84.8; +1.2% | Helped |
| Use 6 × 6 rather than 3 × 3 patches (ablation) | TSViT · Germany | mIoU: 83.6 -> 79.6 | Hurt |
| Use TSViT rather than the strongest reported baseline | Semantic segmentation models · Germany | mIoU: 77.1 -> 84.8 | Helped |
| Use TSViT rather than the strongest reported baseline | Semantic segmentation models · PASTIS | mIoU: 62.4 -> 65.1 | Helped |
| Use TSViT rather than the strongest reported baseline | Semantic segmentation models · T31TFM-1618 | mIoU: 58.8 -> 63.1 | Helped |
| Use TSViT rather than the strongest reported baseline | Object classification models · Germany | mAcc: 84.3 -> 88.1 | Helped |
| Use TSViT rather than the strongest reported baseline | Object classification models · PASTIS | mAcc: 71.8 -> 75.5 | Helped |
| Use TSViT rather than the strongest reported baseline | Object classification models · T31TFM-1618 | mAcc: 71.7 -> 74.2 | Helped |
Closest papers by meaning
- xT: Nested Tokenization for Larger Context in Large Images (ICML 2024 · similarity 0.86)
- TEOChat: A Large Vision-Language Assistant for Temporal Earth Observation Data (ICLR 2025 · similarity 0.85)
- SATA: Spatial Autocorrelation Token Analysis for Enhancing the Robustness of Vision Transformers (CVPR 2025 · similarity 0.85)
- Exact: Exploring Space-Time Perceptive Clues for Weakly Supervised Satellite Image Time Series Semantic Segmentation (CVPR 2025 · similarity 0.85)
- SatCLIP: Global, General-Purpose Location Embeddings with Satellite Imagery (AAAI 2025 · similarity 0.84)
- SatSynth: Augmenting Image-Mask Pairs through Diffusion Models for Aerial Semantic Segmentation (CVPR 2024 · similarity 0.84)
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.