[Python] 圖像處理 – Depthmap 001 – 深度圖介紹與生成

import cv2
from matplotlib import pyplot as plt
# Load the left and right image in grayscale
imgL = cv2.imread('bike_l.png', cv2.IMREAD_GRAYSCALE)
imgR = cv2.imread('bike_r.png', cv2.IMREAD_GRAYSCALE)
# Create a Block Matching process
stereo = cv2.StereoBM.create(numDisparities=0, blockSize=9)
# Run and generate depthmap by computing the spatial differences
disparity = stereo.compute(imgL, imgR)
# Display the image in gray colourmap
plt.imshow(disparity, 'gray')
plt.axis('off')
plt.show()

import cv2
import torch
from matplotlib import pyplot as plt

# Load the BGR image
image = cv2.imread('demo.png')
# Turn it into RGB
imageRGB = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)

# Load the MiDaS Model and decide if use GPU or CPU
midas = torch.hub.load("intel-isl/MiDaS", "DPT_Large")
device = torch.device(
    "cuda") if torch.cuda.is_available() else torch.device("cpu")
midas.to(device)
midas.eval()

# Transform the image for input to MiDaS
midas_transforms = torch.hub.load("intel-isl/MiDaS", "transforms")
transform = midas_transforms.dpt_transform if hasattr(
    midas_transforms, 'dpt_transform') else midas_transforms.default_transform
input_batch = transform(imageRGB).to(device)

# Predict depth map
with torch.no_grad():
    prediction = midas(input_batch)
    prediction = torch.nn.functional.interpolate(prediction.unsqueeze(
        1), size=imageRGB.shape[:2], mode="bicubic", align_corners=False).squeeze()

# Store the result
depth_map = prediction.cpu().numpy()

# Show the grayscale of depthmap
plt.imshow(depth_map, 'gray')
plt.axis('off')
plt.show()