-
Notifications
You must be signed in to change notification settings - Fork 1
/
region_metrics.py
135 lines (114 loc) · 6.74 KB
/
region_metrics.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
from shapely.geometry import Point, LineString, LinearRing, MultiPolygon, Polygon
import numpy as np
from fastprogress import progress_bar
import geopandas as gpd
import pandas as pd
from metrics import *
from shapely.ops import voronoi_diagram as svd
def LinearRingDense(linear_ring, step=1):
xy=[]
for i_vertex, (x, y) in enumerate(linear_ring.coords[:-1]):
xy.append((x,y))
segment = LineString([(x,y),linear_ring.coords[i_vertex+1]])
n_additional_vertex = int(segment.length/step)
segment_step = segment.length/(n_additional_vertex+1) # +1
if n_additional_vertex>0:
for delta in np.arange(0, segment.length, segment_step):
new_point = segment.interpolate(delta)
xy.append((new_point.x,new_point.y))
return LinearRing(xy)
def PolygonDense(polygon, step):
exterior_dense = LinearRingDense(polygon.exterior, step)
holes_dense =[LinearRingDense(interior, step) for interior in polygon.interiors]
return Polygon(exterior_dense, holes=holes_dense)
def MultiPolygonDense(multi_polygon, step):
dense_parts = [PolygonDense(polygon,step) for polygon in multi_polygon]
return MultiPolygon(dense_parts)
def compute_voronoi_polygons(gdf):
# Increase the density of points per edge to better define voronoi regions
point_gdfs = []
for i, r in progress_bar(gdf.iterrows(), total=gdf.shape[0]):
points = list(zip(*PolygonDense(r.geometry, 1).exterior.coords.xy))
point_gdf = gpd.GeoDataFrame(geometry=[Point(p) for p in points], crs=gdf.crs)
point_gdf['ref_poly'] = i # Save reference polygon information
point_gdfs.append(point_gdf)
point_gdfs = pd.concat(point_gdfs)
# Create voronoi regions
smp = point_gdfs.unary_union
smp_vd = svd(smp)
gs = gpd.GeoSeries([smp_vd]).explode(index_parts=True)
gdf_vd_primary = gpd.geodataframe.GeoDataFrame(geometry=gs, crs=gdf.crs)
gdf_vd_primary.reset_index(drop=True)
gdf['voronoi'] = gpd.sjoin(gdf_vd_primary, point_gdfs, how='inner', predicate='intersects').groupby('ref_poly').apply(lambda x : x.unary_union)
return gdf
def compute_object_metrics(eval_y_true_shp, level_1, voronoi_shp, y_pred_shp, buffer=None, buffer_kind='outer'):
"""
Compute metrics for a given reference polygon (eval_y_true_shp) and its corresponding predicted polygon (eval_y_pred_shp)
within an evaluation region (eval_voronoi_shp). Metrics can be relaxed applying a tolerance buffer (buffer).
"""
# Get corresponding voronoi polygon and prediction polygon
eval_voronoi_shp = voronoi_shp[voronoi_shp.level_1 == level_1].unary_union
eval_y_pred_shp = y_pred_shp[y_pred_shp.intersects(eval_voronoi_shp)].unary_union
m = {
'iou_pos': 0., 'dice_pos': 0., 'ppv_pos': 0., 'tpr_pos': 0., 'tnr_pos': 0., 'fpr_pos': 0., 'fnr_pos': 0., 'auc_pos': 0., 'kappa_pos': 0., 'mcc_pos': 0.,
'eval_y_true_shp_pos': eval_y_true_shp, 'eval_y_pred_shp_pos': None, 'eval_buffer_shp_pos': None,
'iou_neg': 0., 'dice_neg': 0., 'ppv_neg': 0., 'tpr_neg': 0., 'tnr_neg': 0., 'fpr_neg': 0., 'fnr_neg': 0., 'auc_neg': 0., 'kappa_neg': 0., 'mcc_neg': 0.,
'eval_y_true_shp_neg': None, 'eval_y_pred_shp_neg': None, 'eval_buffer_shp_neg': None,
'eval_voronoi_shp': eval_voronoi_shp,
'class_weights': None,
'area': eval_y_true_shp.area
}
if eval_y_pred_shp and eval_y_pred_shp.area > 0: # Check if eval_y_pred_shp is not Nonetype due to empty intersection
# Clip predicted polygon and reference ground truth with evaluated voronoi polygon
inner_eval_y_true_shp = eval_y_true_shp.intersection(eval_voronoi_shp)
inner_eval_y_pred_shp = eval_y_pred_shp.intersection(eval_voronoi_shp)
inner_eval_not_y_true_shp = eval_voronoi_shp.difference(inner_eval_y_true_shp)
inner_eval_not_y_pred_shp = eval_voronoi_shp.difference(inner_eval_y_pred_shp)
if inner_eval_y_pred_shp and inner_eval_y_pred_shp.area > 0:
# Buffer
buff = get_buffer_shp(gpd.GeoDataFrame(geometry=[inner_eval_y_true_shp], crs=voronoi_shp.crs), buffer, kind=buffer_kind) if buffer else None
# Compute TPs, TNs, FPs and FNs
tp_pos, tn_pos, fp_pos, fn_pos = compute_confusion_matrix_shp(
gpd.GeoDataFrame({'geometry': [inner_eval_y_pred_shp]}, crs=y_pred_shp.crs),
gpd.GeoDataFrame({'geometry': [inner_eval_y_true_shp]}, crs=y_pred_shp.crs),
buffer=buff,
boundary=eval_voronoi_shp
)
tp_neg, tn_neg, fp_neg, fn_neg = compute_confusion_matrix_shp(
gpd.GeoDataFrame({'geometry': [inner_eval_not_y_pred_shp]}, crs=y_pred_shp.crs),
gpd.GeoDataFrame({'geometry': [inner_eval_not_y_true_shp]}, crs=y_pred_shp.crs),
buffer=buff,
boundary=eval_voronoi_shp
)
class_weights = np.array([inner_eval_y_true_shp.area, inner_eval_not_y_true_shp.area])
class_weights /= class_weights.sum()
m.update({
'iou_pos': iou(tp_pos, tn_pos, fp_pos, fn_pos),
'dice_pos': dice(tp_pos, tn_pos, fp_pos, fn_pos),
'ppv_pos': ppv(tp_pos, tn_pos, fp_pos, fn_pos),
'tpr_pos': tpr(tp_pos, tn_pos, fp_pos, fn_pos),
'tnr_pos': tnr(tp_pos, tn_pos, fp_pos, fn_pos),
'fpr_pos': fpr(tp_pos, tn_pos, fp_pos, fn_pos),
'fnr_pos': fnr(tp_pos, tn_pos, fp_pos, fn_pos),
'auc_pos': auc(tp_pos, tn_pos, fp_pos, fn_pos),
'kappa_pos': kappa(tp_pos, tn_pos, fp_pos, fn_pos),
'mcc_pos': mcc(tp_pos, tn_pos, fp_pos, fn_pos),
'eval_y_true_shp_pos': inner_eval_y_true_shp,
'eval_y_pred_shp_pos': inner_eval_y_pred_shp,
'eval_buffer_shp_pos': buff,
'iou_neg': iou(tp_neg, tn_neg, fp_neg, fn_neg),
'dice_neg': dice(tp_neg, tn_neg, fp_neg, fn_neg),
'ppv_neg': ppv(tp_neg, tn_neg, fp_neg, fn_neg),
'tpr_neg': tpr(tp_neg, tn_neg, fp_neg, fn_neg),
'tnr_neg': tnr(tp_neg, tn_neg, fp_neg, fn_neg),
'fpr_neg': fpr(tp_neg, tn_neg, fp_neg, fn_neg),
'fnr_neg': fnr(tp_neg, tn_neg, fp_neg, fn_neg),
'auc_neg': auc(tp_neg, tn_neg, fp_neg, fn_neg),
'kappa_neg': kappa(tp_neg, tn_neg, fp_neg, fn_neg),
'mcc_neg': mcc(tp_neg, tn_neg, fp_neg, fn_neg),
'eval_y_true_shp_neg': inner_eval_not_y_true_shp,
'eval_y_pred_shp_neg': inner_eval_not_y_pred_shp,
'eval_buffer_shp_neg': buff,
'class_weights': class_weights,
})
return m