Modes other than RETR_FLOODFILL and RETR_CCOMP support only CV_8UC1 images

[johnlockejrr]

I try to get the contours of the detected objects in an image but I get this unspecified error:

0: 640x480 4 persons, 1 bus, 1 skateboard, 74.5ms
Speed: 4.2ms preprocess, 74.5ms inference, 63.3ms postprocess per image at shape (1, 3, 640, 480)
Traceback (most recent call last):
  File "/home/incognito/yolov8/inferemce.py", line 32, in <module>
    contours, _ = cv2.findContours(np.array(resized, np.uint8), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
cv2.error: OpenCV(4.10.0) /io/opencv/modules/imgproc/src/contours_new.cpp:327: error: (-2:Unspecified error) in function 'static std::shared_ptr<ContourScanner_> ContourScanner_::create(cv::Mat, int, int, cv::Point)'
> Modes other than RETR_FLOODFILL and RETR_CCOMP support only CV_8UC1 images (expected: 'img.type() == CV_8UC1'), where
>     'img.type()' is 3832 (CV_8UC480)
> must be equal to
>     'CV_8UC1' is 0 (CV_8UC1)

My code:

import cv2
from matplotlib import pyplot as plt
import numpy as np
import random

# load a pretrained model yolov8x, yolov8x-seg(with segmentation), yolov8x-cls(with classification)
model = YOLO("yolov8n-seg.pt")

img = cv2.imread('bus.jpg') #"https://ultralytics.com/images/bus.jpg"
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

r = model.predict(img)
# r = model(img)

names = model.names
colors = [tuple([random.randint(0, 255) for _ in range(3)]) for _ in range(len(names))]
roi = np.zeros(img.shape).astype(np.int32)
for box, mask in zip(r[0].boxes, r[0].masks):
    # Bounding box & Category
    bbox = box.xyxy.cpu().numpy()[0].astype(np.int32)
    category = int(box.cls.cpu().numpy()[0])
    img = cv2.rectangle(img, (bbox[0], bbox[1]), (bbox[2], bbox[3]), (colors[category]), 5)
    label = names[category] +' '+str(box.conf.cpu().numpy()[0])
    (w, h), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.6, 1)
    # Prints the text.
    img = cv2.rectangle(img, (bbox[0], bbox[1] - 20), (bbox[0] + w, bbox[1]), colors[category], -1)
    img = cv2.putText(img, label, (bbox[0], bbox[1] - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0,0,0), 1)
    # Mask
    m = mask.data.cpu().numpy()
    resized = cv2.resize(m, (img.shape[1], img.shape[0]), interpolation = cv2.INTER_AREA)
    contours, _ = cv2.findContours(np.array(resized, np.uint8), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    cnt = []
    area = 0
    for c in contours:
        a = cv2.contourArea(c)
        if a > area:
            cnt = c
            area = a
    roi = cv2.drawContours(roi, [cnt], -1, colors[category], -1)
blended = ((0.6 * img) + (0.4 * roi)).astype("uint8")
plt.imshow(blended)
plt.show()

[UltralyticsAssistant]

👋 Hello @johnlockejrr, thank you for reaching out about your issue with contour detection 🚀! This is an automated response, but rest assured, an Ultralytics engineer will assist you soon.

The error you're encountering indicates a mismatch in the image type expected by cv2.findContours. This function requires a single-channel (grayscale) image of type CV_8UC1.

Here's how you can adjust your code to fix this issue:

Solution:

Ensure the resized mask image is in a single-channel format before calling cv2.findContours:

# Assuming 'resized' needs to be a binary image (0 and 255)
# Convert the mask to a binary image
resized = (resized > 0.5).astype(np.uint8) * 255

Updated Code:

# Mask
m = mask.data.cpu().numpy()
resized = cv2.resize(m, (img.shape[1], img.shape[0]), interpolation = cv2.INTER_AREA)

# Convert resized mask to binary image
resized_binary = (resized > 0.5).astype(np.uint8) * 255

contours, _ = cv2.findContours(resized_binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
cnt = []
area = 0
for c in contours:
    a = cv2.contourArea(c)
    if a > area:
        cnt = c
        area = a
roi = cv2.drawContours(roi, [cnt], -1, colors[category], -1)

Tips:

1. Verify the mask is in the correct format: Ensure the mask is binary before finding contours.
2. Use OpenCV documentation: Check findContours for more details.
3. Optimize contour processing: Consider filters like area thresholding to remove noise if needed.

If you have more questions or need further clarification, feel free to provide additional details.

Join our community for more support:

• Chat on Discord 🎧
• Discuss on Discourse
• Engage on our Subreddit

Happy coding! 🖥️

[johnlockejrr]

Same error with your code:

0: 640x480 4 persons, 1 bus, 1 skateboard, 51.3ms
Speed: 4.1ms preprocess, 51.3ms inference, 65.6ms postprocess per image at shape (1, 3, 640, 480)
Traceback (most recent call last):
  File "/home/incognito/yolov8/inference.py", line 36, in <module>
    contours, _ = cv2.findContours(resized_binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
cv2.error: OpenCV(4.10.0) /io/opencv/modules/imgproc/src/contours_new.cpp:327: error: (-2:Unspecified error) in function 'static std::shared_ptr<ContourScanner_> ContourScanner_::create(cv::Mat, int, int, cv::Point)'
> Modes other than RETR_FLOODFILL and RETR_CCOMP support only CV_8UC1 images (expected: 'img.type() == CV_8UC1'), where
>     'img.type()' is 3832 (CV_8UC480)
> must be equal to
>     'CV_8UC1' is 0 (CV_8UC1)

[glenn-jocher]

@johnlockejrr ensure the resized_binary image is a single-channel binary image. You can achieve this by verifying the mask conversion step to uint8 and ensuring it's properly thresholded. If the issue persists, double-check the mask processing logic.

[glenn-jocher]

@johnlockejrr you don't need any of that code to get the contour coordintes, just results[0].masks.xy.