Update report (#94)

* update release report

* update leaderboard

* remove yolo

Author: bw4sz

data_prep/package_datasets.py

@@ -91,6 +91,25 @@ def convert_to_shapely(value):
     return datasets
 
 
+def filter_degenerate_polygons(datasets):
+    """Filter out polygons that would create invalid bounding boxes (zero width or height)."""
+    def convert_to_shapely(value):
+        if type(value) == str:
+            return shapely.wkt.loads(value)
+        elif isinstance(value, BaseGeometry):
+            return value
+        else:
+            raise ValueError(f"Invalid geometry type: {type(value)}")
+    
+    shapely_geometries = gpd.GeoSeries(datasets["polygon"].apply(convert_to_shapely))
+    bounds = shapely_geometries.bounds
+    
+    # Filter out polygons where xmin == xmax or ymin == ymax
+    valid_mask = (bounds["minx"] != bounds["maxx"]) & (bounds["miny"] != bounds["maxy"])
+    
+    return datasets.loc[valid_mask].copy()
+
+
 def create_directories(base_dir, dataset_type):
     """Create directories for the dataset."""
     os.makedirs(f"{base_dir}{dataset_type}_{version}/images", exist_ok=True)
@@ -493,6 +512,9 @@ def run(version, base_dir, debug=False):
     TreePoints_datasets = process_geometry_columns(TreePoints_datasets, "point")
     TreePolygons_datasets = process_geometry_columns(TreePolygons_datasets, "polygon")
 
+    # Remove degenerate polygons (those that would create invalid bounding boxes)
+    TreePolygons_datasets = filter_degenerate_polygons(TreePolygons_datasets)
+
     # Copy images
     copy_images(TreeBoxes_datasets, base_dir, "TreeBoxes")
     copy_images(TreePoints_datasets, base_dir, "TreePoints")

docs/examples/baseline_polygons.py

@@ -8,7 +8,7 @@
 import torch
 
 from deepforest import main as df_main
-from deepforest.utilities import read_file
+from deepforest.utilities import read_file, format_geometry
 from deepforest.visualize import plot_results
 
 from milliontrees import get_dataset
@@ -61,20 +61,20 @@ def format_deepforest_predictions(
                                                    "images")
             formatted_pred["image_path"] = basename
         else:
-            pred.root_dir = os.path.join(dataset._data_dir._str, "images")
-            pred["image_path"] = basename
+            formatted_pred = format_geometry(pred)
+            formatted_pred.root_dir = os.path.join(dataset._data_dir._str,
+                                                   "images")
+            formatted_pred["image_path"] = basename
 
             y_pred = {
                 "y": torch.tensor(
-                    pred[["xmin", "ymin", "xmax", "ymax"]].values.astype(
+                    formatted_pred[["xmin", "ymin", "xmax", "ymax"]].values.astype(
                         "float32")),
-                "labels": torch.tensor(pred.label.values.astype(np.int64)),
+                "labels": torch.tensor(formatted_pred.label.values.astype(np.int64)),
                 "scores": torch.tensor(
-                    pred.score.values.astype("float32")),
+                    formatted_pred.score.values.astype("float32")),
             }
 
-            formatted_pred = pred
-
         batch_y_pred.append(y_pred)
         formatted_predictions.append(formatted_pred)
 
@@ -94,8 +94,8 @@ def plot_eval_result(
 
     # Ground truth
     gt_df = read_file(
-        pd.DataFrame(image_targets["y"].numpy(),
-                     columns=["xmin", "ymin", "xmax", "ymax"]))
+        pd.DataFrame(image_targets["bboxes"],
+                     columns=["xmin", "ymin", "xmax", "ymax"]), label="Tree")
     gt_df["label"] = "Tree"
 
     # Predictions
@@ -107,8 +107,8 @@ def plot_eval_result(
     image = image_tensor.permute(1, 2, 0).numpy() * 255
 
     # Simple recall example for logging
-    recall = dataset.metrics["recall"]._recall(image_targets["y"],
-                                               y_pred.get("y",
+    recall = dataset.metrics["recall"]._recall(image_targets["bboxes"],
+                                               y_pred.get("bboxes",
                                                           torch.zeros(
                                                               (0, 4))),
                                                iou_threshold=0.3)
@@ -159,6 +159,7 @@ def main():
     # Load model
     model = df_main.deepforest()
     model.load_model("weecology/deepforest-tree")
+    model.eval()
 
     # Load dataset
     polygon_dataset = get_dataset("TreePolygons",

docs/examples/torchvision_fasterrcnn_treeboxes.py

@@ -59,26 +59,20 @@ All point sources are used to train and predict all box sources.
 |---|---|---|---|---|
 | DeepForest baseline (`baseline_boxes.py`) | TreeBoxes | TreeBoxes | 0.397 | <small>`uv run python docs/examples/baseline_boxes.py --split-scheme random`</small> |
 | SAM3 (`sam3_boxes.py`) | TreeBoxes | TreeBoxes | 0.165 | <small>`uv run python docs/examples/sam3_boxes.py --device cuda --split-scheme random --hf-token $HF_TOKEN`</small> |
-| Torchvision Faster R-CNN (`torchvision_fasterrcnn_treeboxes.py`) | TreeBoxes | TreeBoxes |  | <small>`uv run python docs/examples/torchvision_fasterrcnn_treeboxes.py --split_scheme random`</small> |
-| Ultralytics YOLO (`yolo_treeboxes.py`) | TreeBoxes | TreeBoxes |  | <small>`uv run python docs/examples/yolo_treeboxes.py --split_scheme random --weights yolov8n.pt`</small> |
 
 ### Zero-shot
 
 | Model | Task | Dataset | Avg Recall | Script |
 |---|---|---|---|---|
 | DeepForest baseline (`baseline_boxes.py`) | TreeBoxes | TreeBoxes | 0.534 | <small>`uv run python docs/examples/baseline_boxes.py --split-scheme zeroshot`</small> |
 | SAM3 (`sam3_boxes.py`) | TreeBoxes | TreeBoxes | 0.208 | <small>`uv run python docs/examples/sam3_boxes.py --device cuda --split-scheme zeroshot --hf-token $HF_TOKEN`</small> |
-| Torchvision Faster R-CNN (`torchvision_fasterrcnn_treeboxes.py`) | TreeBoxes | TreeBoxes |  | <small>`uv run python docs/examples/torchvision_fasterrcnn_treeboxes.py --split_scheme zeroshot`</small> |
-| Ultralytics YOLO (`yolo_treeboxes.py`) | TreeBoxes | TreeBoxes | 0.004 | <small>`uv run python docs/examples/yolo_treeboxes.py --split_scheme zeroshot --weights yolov8n.pt`</small> |
 
 ### Cross-geometry
 
 | Model | Task | Dataset | Avg Recall | Script |
 |---|---|---|---|---|
 | DeepForest baseline (`baseline_boxes.py`) | TreeBoxes | TreeBoxes | 0.000 | <small>`uv run python docs/examples/baseline_boxes.py --split-scheme crossgeometry`</small> |
 | SAM3 (`sam3_boxes.py`) | TreeBoxes | TreeBoxes | 0.000 | <small>`uv run python docs/examples/sam3_boxes.py --device cuda --split-scheme crossgeometry --hf-token $HF_TOKEN`</small> |
-| Torchvision Faster R-CNN (`torchvision_fasterrcnn_treeboxes.py`) | TreeBoxes | TreeBoxes |  | <small>`uv run python docs/examples/torchvision_fasterrcnn_treeboxes.py --split_scheme crossgeometry`</small> |
-| Ultralytics YOLO (`yolo_treeboxes.py`) | TreeBoxes | TreeBoxes | 0.000 | <small>`uv run python docs/examples/yolo_treeboxes.py --split_scheme crossgeometry --weights yolov8n.pt`</small> |
 
 ## TreePolygons