# RSRefSeg Model Checkpoints

Trained SigLIP2 + SAM checkpoints for referring expression segmentation on aerial imagery from **[The Aerial-D Dataset](https://huggingface.co/datasets/luisml77/aerial-d)**.

## Models

- **`rsrefseg_combined.pt`** — Trained on 5 datasets (Aerial-D + RefSegRS + RRSIS-D + NWPU-Refer + Urban1960SatSeg). Uses RSRefSeg-L with `facebook/sam-vit-large`.
- **`rsrefseg_aerial-d.pt`** — Trained exclusively on Aerial-D. Uses RSRefSeg-Base with `facebook/sam-vit-base`.

## Usage

```python
# Load and test with the codebase
from model import SigLipSamSegmentator

model = SigLipSamSegmentator(checkpoint_path="rsrefseg_combined.pt")
mask = model.segment(image, "the building in the top left")
```

See training/evaluation code at **[GitHub](https://github.com/luisml77/aerialseg)**.

## Links

- **[📝 Paper](https://arxiv.org/abs/2512.07338)** - arXiv preprint
- **[📊 Aerial-D Dataset](https://huggingface.co/datasets/luisml77/aerial-d)** - Aerial-D dataset
- **[💻 Code](https://github.com/luisml77/aerialseg)** - Training and evaluation scripts
- **[🌐 Project](https://luispl77.github.io/aerial-d/)** - Project page
- **[📦 Complete Collection](https://huggingface.co/collections/luisml77/aerial-d-68a17e2431daebb96218edce)** - All Aerial-D artifacts