Add regression test for Hub model labels

tests/test_hub_models.py

@@ -0,0 +1,39 @@
+import pytest
+import numpy as np
+from deepforest import main
+
+MODELS_TO_TEST = [
+    ("weecology/deepforest-bird", "Bird"),
+    # ("weecology/everglades-nest-detection", "Nest"),
+    ("weecology/deepforest-tree", "Tree"),
+    ("weecology/deepforest-livestock", "Livestock")
+]
+
+@pytest.mark.parametrize("model_name, expected_label", MODELS_TO_TEST)
+def test_hub_model_labels(model_name, expected_label):
+    """
+    Regression test for #1276: Ensure models loaded from Hub
+    return their specific labels, not the default 'Tree'.
+    """
+    m = main.deepforest()
+
+    m.load_model(model_name=model_name)
+
+    assert expected_label in m.label_dict.keys(), \
+        f"Model {model_name} label_dict {m.label_dict} does not contain '{expected_label}'"
+
+    white_img = np.full((2048, 2048, 3), 255, dtype=np.uint8)
+
+    m.config.score_thresh = 0.01
+    m.model.score_thresh = 0.01
+
+    res = m.predict_tile(
+        image=white_img,
+        patch_size=128,
+        patch_overlap=0,
+        iou_threshold=0.1
+    )
+
+    if res is not None and not res.empty:
+        assert (res["label"] == expected_label).all(), \
+            f"Model {model_name} predicted wrong labels: {res['label'].unique()}"