data_io.py

import numpy as np
import cv2
import os


class CarlaDataset(object):
    def __init__(self, data_root='/data3/Carla_stereo'):
        pass
        self.data_root = data_root
        fov = 90        # default setting for carla
        self.height = 1080
        self.width = 1920
        self.focal_length = self.width / (2.0 * np.tan(fov * np.pi / 360.0))
        self.baseline_left_to_right = 2.0       # in terms of meter
        self.baseline_left_to_middle = 1.0
        self.baseline_right_to_middle = 1.0
        self.loc_idx = ['Middle', 'Left', 'Right']
        self.sensor_idx = ['NormalizedDepth', 'RawDepth', 'RGB']
        self.data_dict = {}

    def read_data(self):
        for loc in self.loc_idx:
            for sensor in self.sensor_idx:
                sensor_name = f'{loc}_{sensor}'
                sensor_data = os.listdir(os.path.join(self.data_root, sensor_name))
                self.data_dict['sensor_name'] = sensor_data

    def get_intrinsics(self, fov, h, w):
        pass

    def read_depth(self, raw_depth, norm_depth):
        '''
        raw_depth: [H, W, 3] in RGB format
        '''
        depth = cv2.imread(raw_depth)
        import pdb; pdb.set_trace()
        depth_value = (depth[..., 0] + depth[..., 1] * 256 + depth[..., 2] * (256**2)) / (256**3 - 1)
        norm_depth = cv2.imread(norm_depth)
        valid_mask = norm_depth[..., 0] != 255
        cv2.imwrite('./valid_mask.png', (valid_mask*255).astype(np.uint8))
        return depth_value, valid_mask

if __name__ == '__main__':
    carla_dataset = CarlaDataset()
    raw_depth = 'Middle_RawDepth/000005.png'
    norm_depth = 'Middle_NormalizedDepth/000005.png'
    carla_dataset.read_depth(raw_depth, norm_depth)