load_uoaissim_depth.py

gistfile1.txt

import imageio
import numpy as np
import pyfastnoisesimd as fns # pip install pyfastnoisesimd
import matplotlib.pyplot as plt

def perlin_noise(frequency, width, height):

    noise = fns.Noise()
    noise.NoiseType = 2 # perlin noise
    noise.frequency = frequency
    result = noise.genAsGrid(shape=[height, width], start=[0,0])
    return result

def PerlinDistortion(image):
    """
    """
    height, width = image.shape
    # sample distortion parameters from noise vector
    fx = np.random.uniform(0.0001, 0.1)
    fy = np.random.uniform(0.0001, 0.1)
    fz = np.random.uniform(0.01, 0.1)
    wxy = np.random.uniform(0, 10)
    wz = np.random.uniform(0, 0.005)
    cnd_x = wxy * perlin_noise(fx, width, height)
    cnd_y = wxy * perlin_noise(fy, width, height)
    cnd_z = wz * perlin_noise(fz, width, height)

    cnd_h = np.array(list(range(height)))
    cnd_h = np.expand_dims(cnd_h, -1)
    cnd_h = np.repeat(cnd_h, width, -1)
    cnd_w = np.array(list(range(width)))
    cnd_w = np.expand_dims(cnd_w, 0)
    cnd_w = np.repeat(cnd_w, height, 0)

    noise_cnd_h = np.int16(cnd_h + cnd_x)
    noise_cnd_h = np.clip(noise_cnd_h, 0, (height - 1))
    noise_cnd_w = np.int16(cnd_w + cnd_y)
    noise_cnd_w = np.clip(noise_cnd_w, 0, (width - 1))

    new_img = image[(noise_cnd_h, noise_cnd_w)]
    new_img = new_img = new_img + cnd_z
    return new_img.astype(np.float32)


depth_path = '/home/seung/UOAIS-Sim/train/bin/depth/1452.png'

depth_min = 2500 # 0.25m
depth_max = 15000 # 1.5m
depth = imageio.imread(depth_path).astype(np.float32)

# augmentation for training
depth = PerlinDistortion(depth)
depth[depth > depth_max] = depth_max
depth[depth < depth_min] = depth_min
depth = (depth - depth_min) / (depth_max - depth_min) * 255 
depth = np.expand_dims(depth, -1) # (480, 640, 1), 0~255  --> 네트워크에 넣을 때 255로 다시 나눠줘야함

plt.imshow(depth, cmap='gray')
plt.show()