CLYang617/RemoteSensingChangeDetection-RSCD.HA2F

1import numpy as np2 3 4###################       metrics      ###################5class AverageMeter(object):6    """Computes and stores the average and current value"""7 8    def __init__(self):9        self.initialized = False10        self.val = None11        self.avg = None12        self.sum = None13        self.count = None14 15    def initialize(self, val, weight):16        self.val = val17        self.avg = val18        self.sum = val * weight19        self.count = weight20        self.initialized = True21 22    def update(self, val, weight=1):23        if not self.initialized:24            self.initialize(val, weight)25        else:26            self.add(val, weight)27 28    def add(self, val, weight):29        self.val = val30        self.sum += val * weight31        self.count += weight32        self.avg = self.sum / self.count33 34    def value(self):35        return self.val36 37    def average(self):38        return self.avg39 40    def get_scores(self):41        scores_dict = cm2score(self.sum)42        return scores_dict43 44    def clear(self):45        self.initialized = False46 47 48###################      cm metrics      ###################49class ConfuseMatrixMeter(AverageMeter):50    """Computes and stores the average and current value"""51 52    def __init__(self, n_class):53        super(ConfuseMatrixMeter, self).__init__()54        self.n_class = n_class55 56    def update_cm(self, pr, gt, weight=1):57        """获得当前混淆矩阵，并计算当前F1得分，并更新混淆矩阵"""58        val = get_confuse_matrix(num_classes=self.n_class, label_gts=gt, label_preds=pr)59        self.update(val, weight)60        current_score = cm2F1(val)61        return current_score62 63    def get_scores(self):64        scores_dict = cm2score(self.sum)65        return scores_dict66 67 68def harmonic_mean(xs):69    harmonic_mean = len(xs) / sum((x + 1e-6) ** -1 for x in xs)70    return harmonic_mean71 72 73def cm2F1(confusion_matrix):74    hist = confusion_matrix75    tp = hist[1, 1]76    fn = hist[1, 0]77    fp = hist[0, 1]78    tn = hist[0, 0]79    # recall80    recall = tp / (tp + fn + np.finfo(np.float32).eps)81    # precision82    precision = tp / (tp + fp + np.finfo(np.float32).eps)83    # F1 score84    f1 = 2 * recall * precision / (recall + precision + np.finfo(np.float32).eps)85    return f186 87 88def cm2score(confusion_matrix):89    hist = confusion_matrix90    tp = hist[1, 1]91    fn = hist[1, 0]92    fp = hist[0, 1]93    tn = hist[0, 0]94    # acc95    oa = (tp + tn) / (tp + fn + fp + tn + np.finfo(np.float32).eps)96    # recall97    recall = tp / (tp + fn + np.finfo(np.float32).eps)98    # precision99    precision = tp / (tp + fp + np.finfo(np.float32).eps)100    # F1 score101    f1 = 2 * recall * precision / (recall + precision + np.finfo(np.float32).eps)102    # IoU103    iou = tp / (tp + fp + fn + np.finfo(np.float32).eps)104    # pre105    pre = ((tp + fn) * (tp + fp) + (tn + fp) * (tn + fn)) / (tp + fp + tn + fn) ** 2106    # kappa107    kappa = (oa - pre) / (1 - pre)108    score_dict = {'Kappa': kappa, 'IoU': iou, 'F1': f1, 'OA': oa, 'recall': recall, 'precision': precision, 'Pre': pre}109    return score_dict110 111 112def get_confuse_matrix(num_classes, label_gts, label_preds):113    """计算一组预测的混淆矩阵"""114 115    def __fast_hist(label_gt, label_pred):116        """117        Collect values for Confusion Matrix118        For reference, please see: https://en.wikipedia.org/wiki/Confusion_matrix119        :param label_gt: <np.array> ground-truth120        :param label_pred: <np.array> prediction121        :return: <np.ndarray> values for confusion matrix122        """123        mask = (label_gt >= 0) & (label_gt < num_classes)124        hist = np.bincount(num_classes * label_gt[mask].astype(int) + label_pred[mask],125                           minlength=num_classes ** 2).reshape(num_classes, num_classes)126        return hist127 128    confusion_matrix = np.zeros((num_classes, num_classes))129    for lt, lp in zip(label_gts, label_preds):130        confusion_matrix += __fast_hist(lt.flatten(), lp.flatten())131    return confusion_matrix132