带有热图颜色的图像上的 Plot 边界框注释

Question

I have bounding box annotation data as df like

x           y          width       height
1028.119141,449.497467,667.6237793,62.45513916
737.3796997,352.5843506,297.2002563,49.53338623
730.9188232,399.9641113,148.6001587,58.14785767
671.157959,463.3088684,1020.751343,43.715271
1084.267212,517.9529419,72.13012695,34.972229
729.9176025,359.7701416,303.7685547,37.2722168
1026.231812,456.6779175,661.5819092,50.31750488
662.06073,457.2356262,1032.41156,99.91079712
668.9989624,411.4431152,191.4957275,43.01715088
677.0771484,567.5809937,464.8626709,337.1990356
659.2854614,353.1618652,373.277771,48.56771851
2626.677246,353.6801758,241.1516113,50.53591919
1026.98584,450.5015869,673.2525635,60.3218689
651.710144,349.5532837,405.8690796,57.69006348
230.9644012,29.21221352,277.5276031,70.96037865
724.7105103,359.6932983,328.6080322,26.96270752
689.3218994,427.1000671,957.1763916,165.146637
761.086853,268.6202087,287.673645,158.2123718
1022.51825,453.3616333,766.8713989,93.41070557
2149.582031,223.365921,0.854980469,2.565200806
735.6414185,363.6863098,286.78125,25.07974243
369.3438416,241.1960144,1196.013336,615.5481873
1357.483154,451.0677185,326.9484863,67.52200317
289.6882935,22.07415199,120.0834045,14.12745857
236.7103271,502.4077148,204.8481445,900.6254883
321.4750977,424.7066956,35.31863403,395.5688171
649.9384766,456.4934692,748.755249,169.52948
596.9605103,467.0890808,1193.770203,98.8921814
1010.315857,447.121582,666.9611206,68.6998291
679.3789673,514.437439,492.6141968,48.35473633
674.8457031,411.6835632,211.552124,43.82150269
679.3789673,460.0383301,1016.961121,46.84368896

for an image like this: I have plotted this data on an image using the following python code

import matplotlib.pyplot as plt
from matplotlib.patches import Rectangle
from PIL import Image

# Display the image 
plt.imshow(Image.open('subject_49251684.png'))

# Display Bounding boxes
for row in df.itertuples():
    x = float(row.x)
    y = float(row.y)
    w = float(row.width)
    h = float(row.height)
    plt.gca().add_patch(Rectangle((x,y),w,h,linewidth=1,edgecolor='auto',facecolor='none'))

I got the output something like

Instead of just red bounding boxes, I want to plot all the bounding boxes as a heatmap where overlapping bounding boxes are dense in the color than non-overlapping bounding boxes.

Any help will be much appreciated.

Answer 1

In order to replicate this example, you can use this string s :

s = """1028.119141,449.497467,667.6237793,62.45513916
737.3796997,352.5843506,297.2002563,49.53338623
730.9188232,399.9641113,148.6001587,58.14785767
671.157959,463.3088684,1020.751343,43.715271
1084.267212,517.9529419,72.13012695,34.972229
729.9176025,359.7701416,303.7685547,37.2722168
1026.231812,456.6779175,661.5819092,50.31750488
662.06073,457.2356262,1032.41156,99.91079712
668.9989624,411.4431152,191.4957275,43.01715088
677.0771484,567.5809937,464.8626709,337.1990356
659.2854614,353.1618652,373.277771,48.56771851
2626.677246,353.6801758,241.1516113,50.53591919
1026.98584,450.5015869,673.2525635,60.3218689
651.710144,349.5532837,405.8690796,57.69006348
230.9644012,29.21221352,277.5276031,70.96037865
724.7105103,359.6932983,328.6080322,26.96270752
689.3218994,427.1000671,957.1763916,165.146637
761.086853,268.6202087,287.673645,158.2123718
1022.51825,453.3616333,766.8713989,93.41070557
2149.582031,223.365921,0.854980469,2.565200806
735.6414185,363.6863098,286.78125,25.07974243
369.3438416,241.1960144,1196.013336,615.5481873
1357.483154,451.0677185,326.9484863,67.52200317
289.6882935,22.07415199,120.0834045,14.12745857
236.7103271,502.4077148,204.8481445,900.6254883
321.4750977,424.7066956,35.31863403,395.5688171
649.9384766,456.4934692,748.755249,169.52948
596.9605103,467.0890808,1193.770203,98.8921814
1010.315857,447.121582,666.9611206,68.6998291
679.3789673,514.437439,492.6141968,48.35473633
674.8457031,411.6835632,211.552124,43.82150269
679.3789673,460.0383301,1016.961121,46.84368896"""

One way to see this problem is we can start from a 2d image of zeros, then using filled rectangles, adding to a counter at each pixel. Then, by replacing all 0s with np.nans we can remove them from the final result.

import matplotlib.pyplot as plt
import numpy as np
from PIL import Image

img = Image.open('subject_49251684.png')
rectangles = np.zeros_like(img)[:, :, 0].astype('float32')

for row in s.splitlines():
    x, y, w, h = map(float, row.split(','))
    rectangles[int(y):int(y + h), int(x):int(x + w)] += 1

rectangles[rectangles == 0] = np.nan

# Plotting the main image
plt.imshow(img)

# Overlaying the rectangles on top of original image
# with transparency = 50%
plt.imshow(rectangles, alpha=0.5)

This is the rectangles only on sort of heatmap style:

And this is the output rectangles on top of the original image: