将ndarray从float32转换为uint8会使图像混乱

Question

Following this blog I am trying to apply heatmap to the original image.

However I have a problem converting float32 to uint8. Before converting to uint8 if I save the image with:

plt.imshow(heatmap)
plt.savefig(f'{directory}/heatmap.png', bbox_inches='tight', pad_inches=0.0)

will output this image:

After the line heatmap = np.uint8(255 * heatmap) I save the image again and get the following output:

The shape of image in both cases is (600, 600) . So how can I get 2nd image to be the same as 1st after using uint8?

Answer 1

You have only negative values in heatmap . Since uint8 can only hold numbers between 0 and 255 the line heatmap = np.uint8(255 * heatmap) will only work as intended if the original values of heatmap are lying between 0. and 1. .

Solution:
Rescale the array to the range of [0,255] before casting it to uint8 :

heatmap = np.uint8(np.interp(heatmap, (heatmap.min(), heatmap.max()), (0, 255)))