optgs/evaluation/depth_metrics.py

import numpy as np


def compute_depth_errors(gt, pred):
    """Computation of error metrics between predicted and ground truth depths
    """
    thresh = np.maximum((gt / pred), (pred / gt))
    a1 = (thresh < 1.25).mean()
    a2 = (thresh < 1.25 ** 2).mean()
    a3 = (thresh < 1.25 ** 3).mean()

    rmse = (gt - pred) ** 2
    rmse = np.sqrt(rmse.mean())

    rmse_log = (np.log(gt) - np.log(pred)) ** 2
    rmse_log = np.sqrt(rmse_log.mean())

    abs_rel = np.mean(np.abs(gt - pred) / gt)

    sq_rel = np.mean(((gt - pred) ** 2) / gt)

    return abs_rel, sq_rel, rmse, rmse_log, a1, a2, a3