diff --git a/cv_analysis/utils/structures.py b/cv_analysis/utils/structures.py
index c799df6..b110862 100644
--- a/cv_analysis/utils/structures.py
+++ b/cv_analysis/utils/structures.py
@@ -77,14 +77,6 @@ class Rectangle:
         y2 = y1 + h
         return cls(x1, y1, x2, y2, discrete=discrete)
 
-    @classmethod
-    def from_dict_xywh(cls, xywh: Dict[str, Coord], discrete=True):
-        """Creates a rectangle from a point and a width and height."""
-        x1, y1, w, h = itemgetter("x", "y", "width", "height")(xywh)
-        x2 = x1 + w
-        y2 = y1 + h
-        return cls(x1, y1, x2, y2, discrete=discrete)
-
     def xyxy(self):
         return self.x1, self.y1, self.x2, self.y2
 
diff --git a/cv_analysis/utils/test_metrics.py b/cv_analysis/utils/test_metrics.py
index 9e46e97..91d3676 100644
--- a/cv_analysis/utils/test_metrics.py
+++ b/cv_analysis/utils/test_metrics.py
@@ -1,5 +1,10 @@
+from operator import itemgetter
 from typing import Iterable
+
 import numpy as np
+from funcy import lmap, lpluck
+
+from cv_analysis.utils import lift
 from cv_analysis.utils.structures import Rectangle
 
 
@@ -25,37 +30,28 @@ def compute_page_iou(results_boxes: Iterable[Rectangle], ground_truth_boxes: Ite
     return score
 
 
-def compute_document_score(results_dict, annotation_dict):
+def compute_document_score(result_dict, ground_truth_dicts):
 
-    page_weights = np.array([len(page["cells"]) for page in annotation_dict["pages"]])
-    page_weights = page_weights / sum(page_weights)
+    extract_cells = lambda dicts: lpluck("cells", dicts["pages"])
 
-    scores = []
-    for i in range(len(annotation_dict["pages"])):
-        scores.append(
-            compute_page_iou(
-                map(Rectangle.from_dict_xywh, results_dict["pages"][i]["cells"]),
-                map(Rectangle.from_dict_xywh, annotation_dict["pages"][i]["cells"]),
-            )
-        )
+    cells_per_ground_truth_page, cells_per_result_page = map(extract_cells, (ground_truth_dicts, result_dict))
+    cells_on_page_to_rectangles = lift(rectangle_from_dict)
+    cells_on_pages_to_rectangles = lift(cells_on_page_to_rectangles)
 
-    doc_score = np.average(np.array(scores), weights=page_weights)
+    rectangles_per_ground_truth_page, rectangles_per_result_page = map(
+        cells_on_pages_to_rectangles, (cells_per_ground_truth_page, cells_per_result_page)
+    )
 
-    return doc_score
+    scores = lmap(compute_page_iou, rectangles_per_result_page, rectangles_per_ground_truth_page)
+
+    n_cells_per_page = np.array(lmap(len, cells_per_ground_truth_page))
+    document_score = np.average(scores, weights=n_cells_per_page / n_cells_per_page.sum())
+
+    return document_score
 
 
-"""
-from cv_analysis.utils.test_metrics import *
-
-r1 = Rectangle.from_dict_xywh({'x': 30, 'y': 40, 'width': 50, 'height': 60})
-r2 = Rectangle.from_dict_xywh({'x': 40, 'y': 30, 'width': 55, 'height': 65})
-r3 = Rectangle.from_dict_xywh({'x': 45, 'y': 35, 'width': 45, 'height': 55})
-r4 = Rectangle.from_dict_xywh({'x': 25, 'y': 45, 'width': 45, 'height': 55})
-d1 = {"pages": [{"cells": [r1.json_xywh(), r2.json_xywh()]}]}
-d2 = {"pages": [{"cells": [r3.json_xywh(), r4.json_xywh()]}]}
-
-compute_iou_from_boxes(r1, r2)
-find_max_overlap(r1, [r2, r3, r4])
-compute_page_iou([r1, r2], [r3, r4])
-compute_document_score(d1, d2)
-"""
+def rectangle_from_dict(d):
+    x1, y1, w, h = itemgetter("x", "y", "width", "height")(d)
+    x2 = x1 + w
+    y2 = y1 + h
+    return Rectangle(x1, y1, x2, y2)