File size: 9,798 Bytes
aed6f6f
 
cf7a05c
 
aed6f6f
cf7a05c
 
aed6f6f
cf7a05c
aed6f6f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf7a05c
 
 
aed6f6f
 
 
 
 
 
cf7a05c
 
aed6f6f
 
 
 
 
 
cf7a05c
 
 
aed6f6f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cf7a05c
aed6f6f
 
 
 
 
 
 
 
 
 
cf7a05c
aed6f6f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
"""Compute Satlas semantic, instance, geometric, and global metrics."""

import argparse
import json
from collections import deque
from pathlib import Path

import matplotlib.pyplot as plt
import numpy as np
import torch
from torch.nn import functional as F
import yaml

ROOT = Path(__file__).resolve().parents[1]


def precision_recall_f1(tp, fp, fn):
    precision = tp / max(tp + fp, 1); recall = tp / max(tp + fn, 1)
    return float(precision), float(recall), float(2 * precision * recall / max(precision + recall, 1e-12))


def connected_components(mask):
    """Return 4-connected component coordinate arrays without SciPy."""
    mask = np.asarray(mask, dtype=bool); visited = np.zeros_like(mask); components = []
    height, width = mask.shape
    for row, col in np.argwhere(mask):
        if visited[row, col]: continue
        queue = deque([(int(row), int(col))]); visited[row, col] = True; points = []
        while queue:
            y, x = queue.popleft(); points.append((y, x))
            for ny, nx in ((y - 1, x), (y + 1, x), (y, x - 1), (y, x + 1)):
                if 0 <= ny < height and 0 <= nx < width and mask[ny, nx] and not visited[ny, nx]:
                    visited[ny, nx] = True; queue.append((ny, nx))
        components.append(np.asarray(points, dtype=np.int32))
    return components


def local_peaks(scores, threshold, radius):
    tensor = torch.from_numpy(scores)[None, None]
    pooled = F.max_pool2d(tensor, kernel_size=2 * radius + 1, stride=1, padding=radius)
    candidates = np.argwhere(np.logical_and(scores >= threshold, scores == pooled[0, 0].numpy()))
    if not len(candidates): return np.empty((0, 2), dtype=np.float32)
    order = sorted(candidates, key=lambda point: scores[tuple(point)], reverse=True); selected = []
    for point in order:
        if all(np.sum((point - previous) ** 2) > radius ** 2 for previous in selected): selected.append(point)
    return np.asarray(selected, dtype=np.float32).reshape(-1, 2)


def point_counts(predictions, targets, threshold, nms_radius, tolerance):
    tp = fp = fn = 0
    for scores, target in zip(predictions[:, 0], targets[:, 0]):
        predicted = local_peaks(scores, threshold, nms_radius)
        truth_components = connected_components(target >= 0.5)
        truth = np.asarray([component.mean(0) for component in truth_components], dtype=np.float32).reshape(-1, 2)
        candidates = sorted((float(np.linalg.norm(p - t)), pi, ti) for pi, p in enumerate(predicted)
                            for ti, t in enumerate(truth) if np.linalg.norm(p - t) <= tolerance)
        matched_pred, matched_truth = set(), set()
        for _, pi, ti in candidates:
            if pi not in matched_pred and ti not in matched_truth: matched_pred.add(pi); matched_truth.add(ti)
        tp += len(matched_pred); fp += len(predicted) - len(matched_pred); fn += len(truth) - len(matched_truth)
    return tp, fp, fn


def component_mask(component, shape):
    mask = np.zeros(shape, dtype=bool); mask[component[:, 0], component[:, 1]] = True; return mask


def polygon_counts(predictions, targets, threshold, iou_threshold):
    tp = fp = fn = 0; matched_ious = []
    for scores, target in zip(predictions[:, 0], targets[:, 0]):
        predicted = connected_components(scores >= threshold); truth = connected_components(target >= 0.5)
        pred_masks = [component_mask(component, scores.shape) for component in predicted]
        truth_masks = [component_mask(component, scores.shape) for component in truth]
        candidates = []
        for pi, pred in enumerate(pred_masks):
            for ti, actual in enumerate(truth_masks):
                union = np.logical_or(pred, actual).sum(); iou = np.logical_and(pred, actual).sum() / max(union, 1)
                if iou >= iou_threshold: candidates.append((float(iou), pi, ti))
        matched_pred, matched_truth = set(), set()
        for iou, pi, ti in sorted(candidates, reverse=True):
            if pi not in matched_pred and ti not in matched_truth:
                matched_pred.add(pi); matched_truth.add(ti); matched_ious.append(iou)
        tp += len(matched_pred); fp += len(predicted) - len(matched_pred); fn += len(truth) - len(matched_truth)
    return tp, fp, fn, float(np.mean(matched_ious)) if matched_ious else 0.0


def dilate(mask, radius):
    height, width = mask.shape[-2:]; padded = np.pad(mask, ((0, 0), (radius, radius), (radius, radius)))
    neighborhoods = [padded[:, dy:dy + height, dx:dx + width]
                     for dy in range(2 * radius + 1) for dx in range(2 * radius + 1)
                     if (dy - radius) ** 2 + (dx - radius) ** 2 <= radius ** 2]
    return np.logical_or.reduce(neighborhoods)


def tolerant_line_metrics(prediction, target, threshold, tolerance):
    predicted, truth = prediction[:, 0] >= threshold, target[:, 0] >= 0.5
    matched_pred = np.logical_and(predicted, dilate(truth, tolerance)).sum()
    matched_truth = np.logical_and(truth, dilate(predicted, tolerance)).sum()
    precision = matched_pred / max(predicted.sum(), 1); recall = matched_truth / max(truth.sum(), 1)
    return float(precision), float(recall), float(2 * precision * recall / max(precision + recall, 1e-12))


def mean_iou(prediction, target, classes):
    values = []
    for label in range(classes):
        pred, truth = prediction == label, target == label; union = np.logical_or(pred, truth).sum()
        if union: values.append(np.logical_and(pred, truth).sum() / union)
    return float(np.mean(values))


def main():
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--config", type=Path, default=ROOT / "conf/config.yaml"); parser.add_argument("--prediction", type=Path)
    parser.add_argument("--target", type=Path); parser.add_argument("--output-dir", type=Path); args = parser.parse_args()
    config = yaml.safe_load(args.config.read_text()); evaluation = config["evaluation"]
    pred_path = args.prediction or ROOT / config["paths"]["inference_dir"] / "predictions.npz"
    target_path = args.target or ROOT / config["data"]["root"] / "test.npz"
    if not pred_path.is_file(): raise FileNotFoundError("Run inference before evaluation")
    prediction, target = np.load(pred_path), np.load(target_path)
    if str(prediction["protocol"]) != str(target["protocol"]) or str(target["protocol"]) != config["data"]["protocol"]:
        raise ValueError("prediction, target, and configuration protocols do not match")
    if str(prediction["source"]) != str(target["source"]): raise ValueError("prediction and target sources do not match")
    if prediction["sample_ids"].shape != target["sample_ids"].shape or not np.array_equal(prediction["sample_ids"], target["sample_ids"]):
        raise ValueError("prediction sample_ids do not exactly match target identity/order")
    segmentation = prediction["segmentation"].argmax(1)
    point = precision_recall_f1(*point_counts(prediction["point"], target["point"], evaluation["point_peak_threshold"],
                                             evaluation["point_nms_radius"], evaluation["point_distance_tolerance"]))
    polygon_counts_result = polygon_counts(prediction["polygon"], target["polygon"], evaluation["mask_threshold"],
                                           evaluation["polygon_iou_threshold"])
    polygon = precision_recall_f1(*polygon_counts_result[:3]); line = tolerant_line_metrics(
        prediction["polyline"], target["polyline"], evaluation["mask_threshold"], evaluation["polyline_distance_tolerance"])
    metrics = {
        "segmentation_mIoU": mean_iou(segmentation, target["segmentation"], config["model"]["segmentation_classes"]),
        "regression_MAE": float(np.abs(prediction["regression"] - target["regression"]).mean()),
        "point_precision": point[0], "point_recall": point[1], "point_F1": point[2],
        "polygon_precision": polygon[0], "polygon_recall": polygon[1], "polygon_F1": polygon[2],
        "polygon_matched_IoU": polygon_counts_result[3],
        "polyline_precision": line[0], "polyline_recall": line[1], "polyline_F1": line[2],
        "property_accuracy": float((prediction["property"].argmax(1) == target["property"]).mean()),
        "classification_accuracy": float((prediction["classification"].argmax(1) == target["classification"]).mean()),
        "samples": int(len(segmentation)), "protocol": str(prediction["protocol"]), "checkpoint": str(prediction["checkpoint"]),
        "source": str(prediction["source"]), "identity_verified": True,
    }
    output = args.output_dir or ROOT / config["paths"]["evaluation_dir"]; output.mkdir(parents=True, exist_ok=True)
    (output / "metrics.json").write_text(json.dumps(metrics, indent=2) + "\n")
    last_highres = np.flatnonzero(target["valid_highres_times"][0])[-1]
    last_lowres = np.flatnonzero(target["valid_lowres_times"][0])[-1]
    image = np.clip(target["highres_images"][0, last_highres].transpose(1, 2, 0), 0, 1)
    figure, axes = plt.subplots(2, 4, figsize=(12, 6)); panels = [
        ("NAIP RGB", image, None), ("Segmentation GT", target["segmentation"][0], "tab20"),
        ("Segmentation pred", segmentation[0], "tab20"), ("Regression", prediction["regression"][0, 0], "viridis"),
        ("Point", prediction["point"][0, 0], "magma"), ("Polygon", prediction["polygon"][0, 0], "magma"),
        ("Polyline", prediction["polyline"][0, 0], "magma"),
        ("Sentinel band", target["lowres_images"][0, last_lowres, 0], "viridis")]
    for axis, (title, panel, cmap) in zip(axes.flat, panels): axis.imshow(panel, cmap=cmap); axis.set_title(title); axis.axis("off")
    figure.tight_layout(); figure.savefig(output / "multitask_predictions.png", dpi=150); plt.close(figure)
    print(json.dumps(metrics, indent=2)); print(f"evaluation={output}")


if __name__ == "__main__": main()