utils.py

import re

import numpy as np

import config


def compute_npx_error(prediction, gt, n):
    # computing n-px error
    true_disp = gt
    index = np.argwhere(true_disp > 0).T
    gt[index[0][:], index[1][:], index[2][:]] = np.abs(
        true_disp[index[0][:], index[1][:], index[2][:]] - prediction[index[0][:], index[1][:], index[2][:]])

    correct = (gt[index[0][:], index[1][:], index[2][:]] < n) | \
              (gt[index[0][:], index[1][:], index[2][:]] < true_disp[index[0][:], index[1][:], index[2][:]] * 0.05)

    return 1 - (float(np.sum(correct)) / float(len(index[0])))


def readPFM(file):
    if not isinstance(file, str):
        file = file.numpy().decode()
    file = open(file, 'rb')

    color = None
    width = None
    height = None
    scale = None
    endian = None

    header = file.readline().rstrip().decode()
    if header == 'PF':
        color = True
    elif header == 'Pf':
        color = False
    else:
        raise Exception('Not a PFM file.')

    dim_match = re.match(r'^(\d+)\s(\d+)\s$', file.readline().decode())
    if dim_match:
        width, height = map(int, dim_match.groups())
    else:
        raise Exception('Malformed PFM header.')

    scale = float(file.readline().rstrip().decode())
    if scale < 0:  # little-endian
        endian = '<'
        scale = -scale
    else:
        endian = '>'  # big-endian

    data = np.fromfile(file, endian + 'f')
    shape = (height, width, 3) if color else (height, width)

    data = np.reshape(data, shape)
    data = np.flipud(data)

    # 加入maxdisp限制
    data[data > (config.MAX_DISP - 1)] = config.MAX_DISP - 1
    return data


def mean_std(inputs):
    inputs = np.float32(inputs) / 255.
    inputs[:, :, 0] -= 0.485
    inputs[:, :, 0] /= 0.229
    inputs[:, :, 1] -= 0.456
    inputs[:, :, 1] /= 0.224
    inputs[:, :, 2] -= 0.406
    inputs[:, :, 2] /= 0.225
    return inputs