update model card

README.md

@@ -33,8 +33,6 @@ Prototypical network (ProtoNet) based **1-way 5-shot binary segmentation** on [L
 
 **Key finding:** Pre-training source dominates other factors — self-supervised **SeCo** outperforms all supervised sources, including domain-aligned Million-AID. Δ_pretrain = 0.103 (≈ 2× backbone effect, ≈ 8× fine-tuning effect).
 
-5-fold CV mean ± std across all 90 runs is documented in the [thesis (PDF)](https://github.com/zmgul/few-shot-woodland-segmentation/blob/main/docs/TEZ.pdf).
-
 ## Dataset
 
 [LandCover.ai v1](https://landcover.ai/) — 41 high-resolution aerial images (25–50 cm/pixel) over Poland. 4 labeled land-cover classes + background. Woodland (33.3%) is held out as the **novel class**; building, water, road are **base classes** used for episodic training.
@@ -45,10 +43,13 @@ Prototypical network (ProtoNet) based **1-way 5-shot binary segmentation** on [L
 import torch
 from huggingface_hub import hf_hub_download
 from src.model import ProtoNet  # from GitHub repo
-from src.config import Config
 
-cfg = Config(BACKBONE="resnet50", PRETRAINED="seco", UNFREEZE_FROM="none")
-model = ProtoNet(cfg)
+# Backbone must match the checkpoint; pretrained init is overwritten by load_state_dict
+model = ProtoNet(
+    backbone_name="resnet50",
+    pretrained="imagenet_v1",
+    unfreeze_from="none",
+)
 
 ckpt_path = hf_hub_download(
     repo_id="zmgul/few-shot-woodland-segmentation",