Reading with Pillow

When the TIF file is opened with Pillow, it is automatically recognized as a 32-bit floating point image with one channel.

We can verify this with Image.mode, and the value should be F.

See: https://pillow.readthedocs.io/en/stable/handbook/concepts.html#modes

from pathlib import Path

import matplotlib.pyplot as plt
from PIL import Image

rgb_path = Path("515d7757-dbe1-4bb8-a817-a917043aae9c", "RGB_camera_0.jpg")
depth_path = Path("515d7757-dbe1-4bb8-a817-a917043aae9c", "DepthRaw_camera_0.tif")

rgb = Image.open(rgb_path)
depth = Image.open(depth_path)
depth.mode

'F'

fig, axs = plt.subplots(1, 2, figsize=(20, 8))
axs[0].imshow(rgb)
axs[1].imshow(depth);

Numpy processing

When we convert the PIL Image into a Numpy array, we get an array of shape (h, w), which is typical of single-channel images. We recommend the PIL -> Numpy method when working with depth images as arrays.

The values of the array are the distances of the camera to the pixels, in meters.

import cv2
import numpy as np

depth_arr = np.array(depth)  # convert from PIL to Numpy
depth_arr.shape

(1080, 1920)

# You can also read with OpenCV, which gives the same array, but our tests use
# the Pillow approach so it is more well-maintained.
cv2_arr = cv2.imread(depth_path, cv2.IMREAD_UNCHANGED)
(depth_arr == cv2_arr).all()

np.True_

depth_arr.min(), depth_arr.max()

(np.float32(73.0333), np.float32(1061.6786))

plt.figure(figsize=(16, 9))
# clamp to (0, 300) for a more pronounced visualization
plt.imshow(depth_arr, vmin=0, vmax=300)
plt.colorbar()

for x, y in [
    (1355, 640),
    (660, 430),
    (1750, 200),
]:
    plt.annotate(
        f"{depth_arr[y, x]:.2f} m",
        (x, y),
        textcoords="offset points",
        xytext=(0, 10),
        ha="center",
    )
    plt.plot(x, y, "ro")