能否将matplotlib.imshow的坐标起始点设为0而非-0.5？

Absolutely—you can absolutely modify this behavior! The offset issue you're seeing comes down to matplotlib.imshow's default coordinate system, which aligns pixel centers to integer coordinates instead of pixel corners. Here's how to fix it:

Why the Offset Happens #

By default, imshow places the center of the top-left pixel at coordinate (0.5, 0.5) (when using the default origin='upper' setting). For large images, this 0.5-pixel offset is negligible, but tiny images make it extremely noticeable—hence the misaligned rectangles you're seeing.

Solution 1: Adjust the extent Parameter #

The most straightforward fix is to explicitly set the extent parameter to map pixel corners to integer coordinates. This tells matplotlib to align the top-left corner of the top-left pixel with (0, 0).

For an image with shape (height, width) (rows x columns):

• If using origin='upper' (default, y-axis increases downward), set extent=[0, width, height, 0]
• If using origin='lower' (y-axis increases upward), set extent=[0, width, 0, height]

Example code:

import matplotlib.pyplot as plt
import numpy as np

# Create a tiny test image
tiny_img = np.random.rand(2, 2)

# Plot with default settings (offset will occur)
plt.subplot(1, 2, 1)
plt.imshow(tiny_img)
plt.gca().add_patch(plt.Rectangle((0, 0), 2, 2, fill=False, edgecolor='red', linewidth=2))
plt.title("Default (Offset)")

# Plot with corrected extent
plt.subplot(1, 2, 2)
plt.imshow(tiny_img, extent=[0, tiny_img.shape[1], tiny_img.shape[0], 0])
plt.gca().add_patch(plt.Rectangle((0, 0), 2, 2, fill=False, edgecolor='green', linewidth=2))
plt.title("Corrected (No Offset)")

plt.show()

This will make your rectangles align perfectly with the image boundaries, even for extremely small images.

Solution 2: Use Custom Coordinate Transforms (Advanced) #

If you need more control, you can adjust the transform used for your patches. For example, you can shift the patch coordinates by -0.5 in both x and y to compensate for the default pixel center alignment:

ax = plt.gca()
ax.imshow(tiny_img)
# Shift rectangle to align with pixel corners
rect = plt.Rectangle((-0.5, -0.5), 2, 2, fill=False, edgecolor='blue', linewidth=2)
ax.add_patch(rect)

However, this is less intuitive than setting the extent directly, especially if you're working with multiple patches or dynamic image sizes.

Key Notes #

• Always double-check your origin setting—it directly impacts how the extent maps to the image.
• The extent method works seamlessly with zooming and scaling, so your rectangles will stay aligned even as you adjust the plot view.

内容的提问来源于stack exchange，提问作者Erotemic