depth-map-still.py

import torch
from PIL import Image
import cv2

model_type = (
    "DPT_Large"  # MiDaS v3 - Large     (highest accuracy, slowest inference speed)
)
# model_type = "DPT_Hybrid"   # MiDaS v3 - Hybrid    (medium accuracy, medium inference speed)
# model_type = "MiDaS_small"  # MiDaS v2.1 - Small   (lowest accuracy, highest inference speed)

midas = torch.hub.load("intel-isl/MiDaS", model_type)

device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
midas.to(device)
midas.eval()

midas_transforms = torch.hub.load("intel-isl/MiDaS", "transforms")

if model_type == "DPT_Large" or model_type == "DPT_Hybrid":
    transform = midas_transforms.dpt_transform
else:
    transform = midas_transforms.small_transform

img = cv2.imread("cache/ski-pov.jpg")
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

input_batch = transform(img).to(device)

with torch.no_grad():
    prediction = midas(input_batch)

    prediction = torch.nn.functional.interpolate(
        prediction.unsqueeze(1),
        size=img.shape[:2],
        mode="bicubic",
        align_corners=False,
    ).squeeze()

output = prediction.cpu().numpy()
import matplotlib.pyplot as plt

plt.imshow(output)
plt.show()