Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,8 @@
+# RSRefSeg Checkpoints
+
+This repository hosts two SigLIP2 + SAM checkpoints from the “Generalized Referring Expression Segmentation on Aerial Photos” project:
+
+- **`rsrefseg_combined.pt`** — trained on all five datasets (Aerial-D, RefSegRS, RRSIS-D, NWPU-Refer, Urban1960SatSeg) using the RSRefSeg-L configuration with `facebook/sam-vit-large`.
+- **`rsrefseg_aerial-d.pt`** — trained exclusively on the Aerial-D dataset with the standard RSRefSeg (SAM ViT-Base) setup.
+
+Both checkpoints expect the same tokenizer/text encoder as the training codebase and can be evaluated with `rsrefseg/test.py`.