Document the standard AutoModel(trust_remote_code=True) loading path; update weights path reference

README.md

@@ -46,6 +46,24 @@ python infer.py fundus.png out/
 Writes `out/mask.png` (arteries red, veins blue) and `out/overlay.png`
 (blended with the input). Runs in `bfloat16`, the precision the model was trained in.
 
+The model itself loads the standard Hugging Face way - `transformers.AutoModel`
+with `trust_remote_code=True` downloads the code, config, and weights directly,
+no manual file handling needed:
+
+```python
+import torch
+from transformers import AutoModel
+
+model = AutoModel.from_pretrained(
+    "JonathanFhima/RLAD-SwinV2-Large", trust_remote_code=True
+).to("cuda").to(torch.bfloat16).eval()
+```
+
+`infer.py` still has the actual pre/post-processing (resize, pad, normalize,
+threshold, color-code) — use its `preprocess_image`, `create_vessel_mask`,
+and `blend_overlay` alongside a model loaded this way, or just call
+`infer.load_model(...)` to get both steps at once.
+
 For a full pipeline from a raw fundus image to vasculature biomarkers, see the
 [RLAD + PVBM end-to-end tutorial](https://colab.research.google.com/github/aim-lab/PVBM/blob/main/rlad_pvbm_tutorial.ipynb),
 which feeds this model's output into [PVBM](https://github.com/aim-lab/PVBM).
@@ -75,7 +93,7 @@ Artery/vein Dice score, averaged across the same local, external, and out-of-dis
 ## License
 
 The code (`LUNetv2.py`, `infer.py`) is Apache 2.0. The weights
-(`weights/model.safetensors`) are non-commercial, research use only — the
+(`model.safetensors`) are non-commercial, research use only — the
 training set mixes in subsets that are themselves non-commercial or
 restricted: