0

I am doing a detection with Mask R-CNN of one model available at Train Mask R-CNN for Image Segmentation.

  • Code I
# Load Image
img = cv2.imread("/content/image.jpg")

test_model, inference_config = load_inference_model(1, "/content/mask_rcnn.h5")
image = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# Detect results
r = test_model.detect([image])[0]
colors = random_colors(80)

# Get Coordinates and show it on the image
object_count = len(r["class_ids"])
for i in range(object_count):
    # 1. Mask
    mask = r["masks"][:, :, i]
    contours = get_mask_contours(mask)
    for cnt in contours:
        cv2.polylines(img, [cnt], True, colors[i], 2)
        img = draw_mask(img, [cnt], colors[i])
  • First attempt

After training I tried to change the background color of the image using the code available at Change the Background of Any Image with 5 Lines of Code. Although it is working, it is not worth it for me to use another model to change the background.

  • Second attempt

Using a suggestion that is also available at Can anyone tell me how can I change mask the foreground if I know the color range of background in RGB?. I added the following line of code img2[ ~mask ] = [0,0,0].

This solution works if for the detected object is only one mask, otherwise it will not work so well. Example:

Example

To solve this problem I created an empty list and did the operation inside of the for loop, but it didn't work.

I noticed that in the current way, only the last mask is taken. Is there a way to get all masks? because if there is more than one object in the image this code doesn't work either.

Improve this question
4
  • I can't test it but if you have mask as numpy.array with the same size as image then you can use it to replace pixels in image image[ ~mask ] = [0,0,0] ie. python - Can anyone tell me how can I change mask the foreground if I know the color range of background in RGB? - Stack Overflow Commented Oct 25, 2022 at 10:57
  • @furas your suggestion works if the segmentation of the object is perfect, otherwise it won't work, in the attempts I made only the last mask was taken. Commented Oct 25, 2022 at 15:35
  • if you have more masks then maybe you should do operatioon OR on all masks to create one mask - like full_mask = mask1 | mask2 | mask3 - or in loop full_mask = r["masks"][:, :, 0] and for i in range(1, object_count): full_mask |= r["masks"][:, :, i] Commented Oct 25, 2022 at 15:45
  • It worked perfectly, thanks for helping me one more time. Commented Oct 25, 2022 at 17:59

1 Answer 1

Reset to default
0

I can't test it but if you have mask as numpy.array with the same size as image then you can use it to replace pixels in image

image[ ~mask ] = [0,0,0]

For example

python - Can anyone tell me how can I change mask the foreground if I know the color range of background in RGB? - Stack Overflow

Finding red color in image using Python & OpenCV - Stack Overflow

If you have more masks then you can use operation OR on all masks to create one mask

final_mask = mask1 | mask2 | mask3

In your code it could be

final_mask = r["masks"][:, :, 0] | r["masks"][:, :, 1] | r["masks"][:, :, 2]

or using loop (to make it more universal)

final_mask = r["masks"][:, :, 0] 

for i in range(1, object_count): 
    final_mask |= r["masks"][:, :, i]

If you would have list of masks then you could use

final_mask = masks[0] | masks[1] | masks[2]

or you could use function functools.reduce() (to make it more universal)

from functools import reduce

def or_masks(a, b):
    return a | b

final_mask = reduce(or_masks, masks)

or shorter

from functools import reduce

final_mask = reduce(lambda a,b:a|b, masks)
Improve this answer
Sign up to request clarification or add additional context in comments.

Comments

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.