Convert a YOLOv8 seg model detection result to YOLOv8 label #16658

johnlockejrr · 2024-10-03T09:59:15Z

johnlockejrr
Oct 3, 2024

I try to convert the results of a YOLOv8 seg model to YOLOv8 label format for using in new model training. Is it a valid approach what I do?
Basicly I train my model for manuscript page text segmentation, I ahve two classes "textzone" and "textline", is there a way to print the "textline"s in order like top-down? I know how to do that for bounding boxes but not for masks.

My code so far:

from ultralytics import YOLO
import cv2
import torch
from pathlib import Path
import numpy as np

# Load a pretrained YOLOv8n-seg segmentation model
model = YOLO("./runs/segment/train2/weights/best.pt")

# Run inference on an image
image_path = './3_page-0018.jpg' # Replace with your image file path
results = model(image_path)

result = results[0]

# Create output directory
output_dir = Path("./test_output/")
output_dir.mkdir(parents=True, exist_ok=True)

# Write the original image for reference
cv2.imwrite(str(output_dir / "original_image.jpg"), result.orig_img)

# Prepare to collect YOLO annotations
yolo_annotations = []

# Dimensions of the input image
img_width, img_height = result.orig_img.shape[1], result.orig_img.shape[0]

# Iterate through each detected result
for result in results:
    masks = result.masks.data
    boxes = result.boxes.data

    clss = boxes[:, 5]  # Get class indices

    # Loop through each instance with detected masks
    for i, mask in enumerate(masks):
        class_index = int(clss[i])  # Get the class index for this mask

        # Convert the mask to a binary format and find contours (polygon representation)
        mask_np = mask.cpu().numpy() > 0 # Convert to binary mask
        contours, _ = cv2.findContours(mask_np.astype(np.uint8), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

        # Process each contour to create the YOLO format polygon annotation
        for contour in contours:
            # Start the annotation with the class index
            annotation = f"{class_index}"
            
            # Collect normalized points
            points = []
            for point in contour:
                x, y = point[0]
                # Normalize coordinates by image dimensions
                normalized_x = x / img_width
                normalized_y = y / img_height
                points.append(f"{normalized_x:.6f} {normalized_y:.6f}")

            # Append all normalized points to the annotation
            annotation += ' ' + ' '.join(points)
            yolo_annotations.append(annotation)

# Write YOLO annotations to a text file
annotations_file_path = output_dir / 'yolo_annotations.txt'
with open(annotations_file_path, 'w') as f:
    for annotation in yolo_annotations:
        f.write(annotation + '\n')

print(f"YOLOv8 annotations saved to '{annotations_file_path}'.")

Output:

0 0.229435 0.033637 0.229435 0.036773 0.229032 0.037058 0.229032 0.038769 0.228629 0.039054 0.228629 0.040764 0.229032 0.041049 0.229032 0.041619 0.228629 0.041904 0.228629 0.043044 0.228226 0.043330 0.228226 0.044755 0.228629 0.045040 0.228629 0.047891 0.228226 0.048176 0.228226 0.051026 0.228629 0.051311 0.228629 0.052452 0.229032 0.052737 0.229032 0.053592 0.229435 0.053877 0.229435 0.054162 0.229032 0.054447 0.229032 0.055017 0.228629 0.055302 0.228629 0.056157 0.229032 0.056442 0.229032 0.059293 0.228629 0.059578 0.228629 0.060718 0.228226 0.061003 0.228629 0.061288 0.228629 0.062144 0.228226 0.062429 0.228226 0.064709 0.227823 0.064994 0.227823 0.065849 0.228226 0.066135 0.228226 0.073546 0.229032 0.074116 0.229032 0.074401 0.229435 0.074686 0.229032 0.074971 0.229032 0.076397 0.229435 0.076682 0.229435 0.078677 0.229032 0.078962 0.229032 0.079818 0.228629 0.080103 0.228629 0.081243 0.228226 0.081528 0.228226 0.085234 0.227823 0.085519 0.227823 0.086089 0.227016 0.086659 0.217742 0.086659 0.217339 0.086944 0.212903 0.086944 0.212500 0.086659 0.209677 0.086659 0.209274 0.086374 0.208871 0.086374 0.208468 0.086089 0.183468 0.086089 0.183065 0.085804 0.182661 0.085804 0.182258 0.086089 0.180242 0.086089 0.179839 0.085804 0.162097 0.085804 0.162097 0.086944 0.163710 0.086944 0.164113 0.086659 0.173790 0.086659 0.174194 0.086374 0.175000 0.086374 0.175403 0.086659 0.204435 0.086659 0.204839 0.086944 0.205242 0.086944 0.205645 0.087229 0.205645 0.088084 0.228629 0.088084 0.228629 0.086944 0.229435 0.086374 0.231048 0.086374 0.231048 0.070410 0.229839 0.070410 0.229435 0.070125 0.229435 0.068985 0.229839 0.068700 0.231048 0.068700 0.231048 0.065849 0.229839 0.065849 0.229435 0.065564 0.229435 0.065279 0.229839 0.064994 0.231048 0.064994 0.231048 0.033637
1 0.162097 0.043900 0.162097 0.049031 0.177419 0.049031 0.177823 0.048746 0.200000 0.048746 0.200403 0.049031 0.208065 0.049031 0.208468 0.048746 0.222984 0.048746 0.223387 0.049031 0.227016 0.049031 0.227823 0.048461 0.229435 0.048461 0.229435 0.043900
1 0.162097 0.074686 0.162097 0.079818 0.164919 0.079818 0.165323 0.080103 0.168548 0.080103 0.168952 0.080388 0.168952 0.081243 0.227419 0.081243 0.227419 0.080388 0.228226 0.079818 0.229435 0.079818 0.229435 0.074686
1 0.162097 0.048461 0.162097 0.055017 0.229435 0.055017 0.229435 0.049886 0.228226 0.049886 0.227419 0.049316 0.227419 0.048461
1 0.162097 0.054162 0.162097 0.059293 0.181855 0.059293 0.182258 0.059578 0.189516 0.059578 0.189919 0.059293 0.227419 0.059293 0.227823 0.059008 0.229435 0.059008 0.229435 0.054162
1 0.163710 0.080388 0.163710 0.085804 0.204839 0.085804 0.205242 0.086089 0.205242 0.086944 0.227419 0.086944 0.227419 0.086089 0.227823 0.085804 0.227823 0.085519 0.228226 0.085234 0.229435 0.085234 0.229435 0.084094 0.228226 0.084094 0.227823 0.083808 0.227823 0.083523 0.228226 0.083238 0.229435 0.083238 0.229435 0.080388
1 0.162097 0.038198 0.162097 0.044755 0.176210 0.044755 0.176210 0.043900 0.176613 0.043615 0.179839 0.043615 0.180242 0.043900 0.180242 0.044755 0.181048 0.044755 0.181048 0.043900 0.181452 0.043615 0.186694 0.043615 0.187097 0.043330 0.189113 0.043330 0.189516 0.043615 0.192339 0.043615 0.192742 0.043900 0.192742 0.044755 0.229435 0.044755 0.229435 0.038198
1 0.162097 0.058723 0.162097 0.064994 0.163710 0.064994 0.164113 0.064709 0.175000 0.064709 0.175403 0.064424 0.216129 0.064424 0.216532 0.064709 0.226613 0.064709 0.227016 0.064424 0.227419 0.064424 0.227419 0.064139 0.227823 0.063854 0.227823 0.063569 0.228226 0.063284 0.229435 0.063284 0.229435 0.060433 0.228226 0.060433 0.227016 0.059578 0.227016 0.058723
1 0.162097 0.070125 0.162097 0.074971 0.165323 0.074971 0.165726 0.074686 0.174597 0.074686 0.175000 0.074971 0.179839 0.074971 0.180242 0.074686 0.197177 0.074686 0.197581 0.074971 0.223790 0.074971 0.224194 0.075257 0.226613 0.075257 0.227016 0.074971 0.227419 0.074971 0.227823 0.074686 0.229435 0.074686 0.229435 0.071836 0.228226 0.071836 0.227823 0.071551 0.227823 0.070125
1 0.162097 0.064424 0.162097 0.070981 0.185081 0.070981 0.185081 0.070125 0.185484 0.069840 0.211290 0.069840 0.211694 0.070125 0.211694 0.070981 0.227016 0.070981 0.227016 0.069840 0.227419 0.069555 0.227419 0.068415 0.227823 0.068130 0.227823 0.064424
1 0.162097 0.033637 0.162097 0.038483 0.164113 0.038483 0.164516 0.038198 0.168548 0.038198 0.168952 0.038483 0.178629 0.038483 0.179032 0.038198 0.181452 0.038198 0.181855 0.038483 0.189516 0.038483 0.189919 0.038198 0.201210 0.038198 0.201613 0.038483 0.226613 0.038483 0.227016 0.038198 0.229435 0.038198 0.229435 0.033637

My second approach:

import torch
import cv2
import numpy as np
from ultralytics import YOLO
from pathlib import Path

# Load YOLOv8 segmentation model
model = YOLO("./runs/segment/train2/weights/best.pt")

# Run inference on an image
results = model('./3_page-0018.jpg')

# Directory to save YOLOv8 labels
output_directory = Path("./output_yolo_labels/")
output_directory.mkdir(parents=True, exist_ok=True)

# Get the first result (assuming one image for simplicity)
result = results[0]
image_height, image_width = result.orig_img.shape[:2]

# Open file to save YOLOv8 format multipoint polygons
txt_file_path = output_directory / "image_0018.txt"
with open(txt_file_path, "w") as yolo_file:
    for mask_idx, (mask, box) in enumerate(zip(result.masks.data, result.boxes)):
        # Get class index from result.boxes.cls attribute
        class_idx = int(result.boxes.cls[mask_idx])

        # Convert the mask to binary format
        mask_2d = mask.cpu().numpy().astype(np.uint8) * 255

        # Find contours (polygons) in the mask
        contours, _ = cv2.findContours(mask_2d, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

        # Loop through each contour (polygon)
        for contour in contours:
            # Normalize each point in the contour (polygon)
            normalized_polygon = []
            for point in contour:
                x, y = point[0]
                x_norm = x / image_width  # Normalize by image width
                y_norm = y / image_height  # Normalize by image height
                normalized_polygon.append(f"{x_norm} {y_norm}")

            # Write the polygon as a line in YOLOv8 TXT format (class index + normalized coordinates)
            yolo_file.write(f"{class_idx} {' '.join(normalized_polygon)}\n")

print(f"YOLOv8 polygon labels saved to {txt_file_path}")

Result:

0 0.22943548387096774 0.033637400228050174 0.22943548387096774 0.03677309007981756 0.22903225806451613 0.037058152793614595 0.22903225806451613 0.03876852907639681 0.22862903225806452 0.03905359179019384 0.22862903225806452 0.04076396807297605 0.22903225806451613 0.04104903078677309 0.22903225806451613 0.04161915621436716 0.22862903225806452 0.0419042189281642 0.22862903225806452 0.043044469783352336 0.22822580645161292 0.043329532497149374 0.22822580645161292 0.04475484606613455 0.22862903225806452 0.04503990877993159 0.22862903225806452 0.04789053591790194 0.22822580645161292 0.04817559863169897 0.22822580645161292 0.05102622576966933 0.22862903225806452 0.05131128848346636 0.22862903225806452 0.052451539338654506 0.22903225806451613 0.05273660205245154 0.22903225806451613 0.053591790193842644 0.22943548387096774 0.05387685290763968 0.22943548387096774 0.05416191562143672 0.22903225806451613 0.05444697833523375 0.22903225806451613 0.05501710376282782 0.22862903225806452 0.05530216647662486 0.22862903225806452 0.056157354618015964 0.22903225806451613 0.056442417331812995 0.22903225806451613 0.059293044469783354 0.22862903225806452 0.059578107183580385 0.22862903225806452 0.06071835803876853 0.22822580645161292 0.06100342075256556 0.22862903225806452 0.0612884834663626 0.22862903225806452 0.062143671607753706 0.22822580645161292 0.062428734321550744 0.22822580645161292 0.06470923603192702 0.22782258064516128 0.06499429874572406 0.22782258064516128 0.06584948688711517 0.22822580645161292 0.0661345496009122 0.22822580645161292 0.07354618015963511 0.22903225806451613 0.07411630558722919 0.22903225806451613 0.07440136830102623 0.22943548387096774 0.07468643101482327 0.22903225806451613 0.0749714937286203 0.22903225806451613 0.07639680729760548 0.22943548387096774 0.0766818700114025 0.22943548387096774 0.07867730900798175 0.22903225806451613 0.07896237172177879 0.22903225806451613 0.07981755986316989 0.22862903225806452 0.08010262257696693 0.22862903225806452 0.08124287343215507 0.22822580645161292 0.0815279361459521 0.22822580645161292 0.08523375142531357 0.22782258064516128 0.08551881413911061 0.22782258064516128 0.08608893956670467 0.22701612903225807 0.08665906499429875 0.21774193548387097 0.08665906499429875 0.21733870967741936 0.08694412770809579 0.2129032258064516 0.08694412770809579 0.2125 0.08665906499429875 0.20967741935483872 0.08665906499429875 0.20927419354838708 0.08637400228050171 0.20887096774193548 0.08637400228050171 0.20846774193548387 0.08608893956670467 0.18346774193548387 0.08608893956670467 0.18306451612903227 0.08580387685290763 0.18266129032258063 0.08580387685290763 0.18225806451612903 0.08608893956670467 0.18024193548387096 0.08608893956670467 0.17983870967741936 0.08580387685290763 0.1620967741935484 0.08580387685290763 0.1620967741935484 0.08694412770809579 0.16370967741935483 0.08694412770809579 0.16411290322580646 0.08665906499429875 0.17379032258064517 0.08665906499429875 0.17419354838709677 0.08637400228050171 0.175 0.08637400228050171 0.17540322580645162 0.08665906499429875 0.20443548387096774 0.08665906499429875 0.20483870967741935 0.08694412770809579 0.20524193548387096 0.08694412770809579 0.2056451612903226 0.08722919042189281 0.2056451612903226 0.08808437856328392 0.22862903225806452 0.08808437856328392 0.22862903225806452 0.08694412770809579 0.22943548387096774 0.08637400228050171 0.2310483870967742 0.08637400228050171 0.2310483870967742 0.07041049030786774 0.22983870967741934 0.07041049030786774 0.22943548387096774 0.0701254275940707 0.22943548387096774 0.06898517673888255 0.22983870967741934 0.06870011402508552 0.2310483870967742 0.06870011402508552 0.2310483870967742 0.06584948688711517 0.22983870967741934 0.06584948688711517 0.22943548387096774 0.06556442417331813 0.22943548387096774 0.0652793614595211 0.22983870967741934 0.06499429874572406 0.2310483870967742 0.06499429874572406 0.2310483870967742 0.033637400228050174
1 0.1620967741935484 0.04389965792474344 0.1620967741935484 0.04903078677309008 0.1774193548387097 0.04903078677309008 0.1778225806451613 0.04874572405929305 0.2 0.04874572405929305 0.20040322580645162 0.04903078677309008 0.20806451612903226 0.04903078677309008 0.20846774193548387 0.04874572405929305 0.22298387096774194 0.04874572405929305 0.22338709677419355 0.04903078677309008 0.22701612903225807 0.04903078677309008 0.22782258064516128 0.04846066134549601 0.22943548387096774 0.04846066134549601 0.22943548387096774 0.04389965792474344
1 0.1620967741935484 0.07468643101482327 0.1620967741935484 0.07981755986316989 0.16491935483870968 0.07981755986316989 0.16532258064516128 0.08010262257696693 0.1685483870967742 0.08010262257696693 0.1689516129032258 0.08038768529076397 0.1689516129032258 0.08124287343215507 0.22741935483870968 0.08124287343215507 0.22741935483870968 0.08038768529076397 0.22822580645161292 0.07981755986316989 0.22943548387096774 0.07981755986316989 0.22943548387096774 0.07468643101482327
1 0.1620967741935484 0.04846066134549601 0.1620967741935484 0.05501710376282782 0.22943548387096774 0.05501710376282782 0.22943548387096774 0.049885974914481185 0.22822580645161292 0.049885974914481185 0.22741935483870968 0.049315849486887116 0.22741935483870968 0.04846066134549601
1 0.1620967741935484 0.05416191562143672 0.1620967741935484 0.059293044469783354 0.18185483870967742 0.059293044469783354 0.18225806451612903 0.059578107183580385 0.18951612903225806 0.059578107183580385 0.18991935483870967 0.059293044469783354 0.22741935483870968 0.059293044469783354 0.22782258064516128 0.059007981755986316 0.22943548387096774 0.059007981755986316 0.22943548387096774 0.05416191562143672
1 0.16370967741935483 0.08038768529076397 0.16370967741935483 0.08580387685290763 0.20483870967741935 0.08580387685290763 0.20524193548387096 0.08608893956670467 0.20524193548387096 0.08694412770809579 0.22741935483870968 0.08694412770809579 0.22741935483870968 0.08608893956670467 0.22782258064516128 0.08580387685290763 0.22782258064516128 0.08551881413911061 0.22822580645161292 0.08523375142531357 0.22943548387096774 0.08523375142531357 0.22943548387096774 0.08409350057012542 0.22822580645161292 0.08409350057012542 0.22782258064516128 0.0838084378563284 0.22782258064516128 0.08352337514253136 0.22822580645161292 0.08323831242873432 0.22943548387096774 0.08323831242873432 0.22943548387096774 0.08038768529076397
1 0.1620967741935484 0.03819840364880274 0.1620967741935484 0.04475484606613455 0.17620967741935484 0.04475484606613455 0.17620967741935484 0.04389965792474344 0.17661290322580644 0.043614595210946405 0.17983870967741936 0.043614595210946405 0.18024193548387096 0.04389965792474344 0.18024193548387096 0.04475484606613455 0.1810483870967742 0.04475484606613455 0.1810483870967742 0.04389965792474344 0.1814516129032258 0.043614595210946405 0.18669354838709679 0.043614595210946405 0.1870967741935484 0.043329532497149374 0.18911290322580646 0.043329532497149374 0.18951612903225806 0.043614595210946405 0.19233870967741937 0.043614595210946405 0.19274193548387097 0.04389965792474344 0.19274193548387097 0.04475484606613455 0.22943548387096774 0.04475484606613455 0.22943548387096774 0.03819840364880274
1 0.1620967741935484 0.05872291904218928 0.1620967741935484 0.06499429874572406 0.16370967741935483 0.06499429874572406 0.16411290322580646 0.06470923603192702 0.175 0.06470923603192702 0.17540322580645162 0.06442417331812998 0.2161290322580645 0.06442417331812998 0.21653225806451612 0.06470923603192702 0.22661290322580646 0.06470923603192702 0.22701612903225807 0.06442417331812998 0.22741935483870968 0.06442417331812998 0.22741935483870968 0.06413911060433296 0.22782258064516128 0.06385404789053592 0.22782258064516128 0.06356898517673888 0.22822580645161292 0.06328392246294184 0.22943548387096774 0.06328392246294184 0.22943548387096774 0.06043329532497149 0.22822580645161292 0.06043329532497149 0.22701612903225807 0.059578107183580385 0.22701612903225807 0.05872291904218928
1 0.1620967741935484 0.0701254275940707 0.1620967741935484 0.0749714937286203 0.16532258064516128 0.0749714937286203 0.16572580645161292 0.07468643101482327 0.17459677419354838 0.07468643101482327 0.175 0.0749714937286203 0.17983870967741936 0.0749714937286203 0.18024193548387096 0.07468643101482327 0.1971774193548387 0.07468643101482327 0.1975806451612903 0.0749714937286203 0.22379032258064516 0.0749714937286203 0.22419354838709676 0.07525655644241733 0.22661290322580646 0.07525655644241733 0.22701612903225807 0.0749714937286203 0.22741935483870968 0.0749714937286203 0.22782258064516128 0.07468643101482327 0.22943548387096774 0.07468643101482327 0.22943548387096774 0.07183580387685291 0.22822580645161292 0.07183580387685291 0.22782258064516128 0.07155074116305588 0.22782258064516128 0.0701254275940707
1 0.1620967741935484 0.06442417331812998 0.1620967741935484 0.0709806157354618 0.18508064516129033 0.0709806157354618 0.18508064516129033 0.0701254275940707 0.18548387096774194 0.06984036488027366 0.21129032258064517 0.06984036488027366 0.21169354838709678 0.0701254275940707 0.21169354838709678 0.0709806157354618 0.22701612903225807 0.0709806157354618 0.22701612903225807 0.06984036488027366 0.22741935483870968 0.06955530216647662 0.22741935483870968 0.06841505131128849 0.22782258064516128 0.06812998859749145 0.22782258064516128 0.06442417331812998
1 0.1620967741935484 0.033637400228050174 0.1620967741935484 0.03848346636259977 0.16411290322580646 0.03848346636259977 0.16451612903225807 0.03819840364880274 0.1685483870967742 0.03819840364880274 0.1689516129032258 0.03848346636259977 0.1786290322580645 0.03848346636259977 0.17903225806451614 0.03819840364880274 0.1814516129032258 0.03819840364880274 0.18185483870967742 0.03848346636259977 0.18951612903225806 0.03848346636259977 0.18991935483870967 0.03819840364880274 0.20120967741935483 0.03819840364880274 0.20161290322580644 0.03848346636259977 0.22661290322580646 0.03848346636259977 0.22701612903225807 0.03819840364880274 0.22943548387096774 0.03819840364880274 0.22943548387096774 0.033637400228050174

Answered by johnlockejrr

Oct 3, 2024

I changed the second script like this as you suggested, I'm not sure the polygons are sorted top down:

import torch
import cv2
import numpy as np
from ultralytics import YOLO
from pathlib import Path

# Load YOLOv8 segmentation model
model = YOLO("./runs/segment/train2/weights/best.pt")

# Run inference on an image
results = model('./3_page-0018.jpg')

# Directory to save YOLOv8 labels
output_directory = Path("./output_yolo_labels/")
output_directory.mkdir(parents=True, exist_ok=True)

# Get the first result (assuming one image for simplicity)
result = results[0]
image_height, image_width = result.orig_img.shape[:2]

# Open file to save YOLOv8 format multipoint polygons
txt_file_path = ou…

View full answer

UltralyticsAssistant · 2024-10-03T09:59:40Z

UltralyticsAssistant
Oct 3, 2024
Maintainer

👋 Hello @johnlockejrr, thank you for sharing your project with us! 🚀

It seems you're working on converting YOLOv8 segmentation results into a label format suitable for training. If you're encountering any issues or seeking validation on your approach, please ensure that you are using the latest version of Ultralytics and provide a minimum reproducible example to help us understand better. Your code snippet is a great start, but an MRE ensures we can see exactly what you're experiencing.

For comprehensive guidance, you might want to check out our Docs where we have sections dedicated to Python and CLI usage examples.

To engage with our community for support or brainstorming, join us on Discord 🎧, or participate in discussions over at Discourse. We also have a Subreddit for broader conversations.

Upgrade

Ensure you are using the latest ultralytics package and all dependencies are up-to-date. Here's how you can upgrade:

pip install -U ultralytics

Environments

YOLO can be run in verified environments with all dependencies preinstalled, and it's always recommended to use these to mitigate configuration issues:

Notebooks with free GPU:

An Ultralytics engineer will be along soon to assist further. Thank you for your patience! 😊

0 replies

glenn-jocher · 2024-10-03T15:19:02Z

glenn-jocher
Oct 3, 2024
Maintainer

@johnlockejrr converting segmentation results to YOLO label format is a valid approach for training. To order "textline" masks top-down, you can sort the contours by their y-coordinate before processing. This will ensure the annotations are in the desired order. If you need further assistance, feel free to ask!

0 replies

johnlockejrr · 2024-10-03T16:40:34Z

johnlockejrr
Oct 3, 2024
Author

I changed the second script like this as you suggested, I'm not sure the polygons are sorted top down:

import torch
import cv2
import numpy as np
from ultralytics import YOLO
from pathlib import Path

# Load YOLOv8 segmentation model
model = YOLO("./runs/segment/train2/weights/best.pt")

# Run inference on an image
results = model('./3_page-0018.jpg')

# Directory to save YOLOv8 labels
output_directory = Path("./output_yolo_labels/")
output_directory.mkdir(parents=True, exist_ok=True)

# Get the first result (assuming one image for simplicity)
result = results[0]
image_height, image_width = result.orig_img.shape[:2]

# Open file to save YOLOv8 format multipoint polygons
txt_file_path = output_directory / "image_0018.txt"

# Collect all the masks and their corresponding bounding boxes
masks_and_boxes = []

for mask_idx, (mask, box) in enumerate(zip(result.masks.data, result.boxes)):
    # Get class index from result.boxes.cls attribute
    class_idx = int(result.boxes.cls[mask_idx])

    # Convert the mask to binary format
    mask_2d = mask.cpu().numpy().astype(np.uint8) * 255

    # Find contours (polygons) in the mask
    contours, _ = cv2.findContours(mask_2d, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

    # Save the mask, class, and its bounding box's topmost y-coordinate for sorting
    topmost_y = min([point[0][1] for contour in contours for point in contour])
    masks_and_boxes.append((mask, class_idx, topmost_y, contours))

# Sort the masks based on the topmost y-coordinate (top-down)
masks_and_boxes.sort(key=lambda x: x[2])

# Process the sorted masks and write to the YOLOv8 TXT file
with open(txt_file_path, "w") as yolo_file:
    for mask, class_idx, topmost_y, contours in masks_and_boxes:
        # Loop through each contour (polygon)
        for contour in contours:
            # Normalize each point in the contour (polygon)
            normalized_polygon = []
            for point in contour:
                x, y = point[0]
                x_norm = x / image_width  # Normalize by image width
                y_norm = y / image_height  # Normalize by image height
                normalized_polygon.append(f"{x_norm} {y_norm}")

            # Write the polygon as a line in YOLOv8 TXT format (class index + normalized coordinates)
            yolo_file.write(f"{class_idx} {' '.join(normalized_polygon)}\n")

print(f"YOLOv8 polygon labels saved to {txt_file_path}")

Result:

0 0.22943548387096774 0.033637400228050174 0.22943548387096774 0.03677309007981756 0.22903225806451613 0.037058152793614595 0.22903225806451613 0.03876852907639681 0.22862903225806452 0.03905359179019384 0.22862903225806452 0.04076396807297605 0.22903225806451613 0.04104903078677309 0.22903225806451613 0.04161915621436716 0.22862903225806452 0.0419042189281642 0.22862903225806452 0.043044469783352336 0.22822580645161292 0.043329532497149374 0.22822580645161292 0.04475484606613455 0.22862903225806452 0.04503990877993159 0.22862903225806452 0.04789053591790194 0.22822580645161292 0.04817559863169897 0.22822580645161292 0.05102622576966933 0.22862903225806452 0.05131128848346636 0.22862903225806452 0.052451539338654506 0.22903225806451613 0.05273660205245154 0.22903225806451613 0.053591790193842644 0.22943548387096774 0.05387685290763968 0.22943548387096774 0.05416191562143672 0.22903225806451613 0.05444697833523375 0.22903225806451613 0.05501710376282782 0.22862903225806452 0.05530216647662486 0.22862903225806452 0.056157354618015964 0.22903225806451613 0.056442417331812995 0.22903225806451613 0.059293044469783354 0.22862903225806452 0.059578107183580385 0.22862903225806452 0.06071835803876853 0.22822580645161292 0.06100342075256556 0.22862903225806452 0.0612884834663626 0.22862903225806452 0.062143671607753706 0.22822580645161292 0.062428734321550744 0.22822580645161292 0.06470923603192702 0.22782258064516128 0.06499429874572406 0.22782258064516128 0.06584948688711517 0.22822580645161292 0.0661345496009122 0.22822580645161292 0.07354618015963511 0.22903225806451613 0.07411630558722919 0.22903225806451613 0.07440136830102623 0.22943548387096774 0.07468643101482327 0.22903225806451613 0.0749714937286203 0.22903225806451613 0.07639680729760548 0.22943548387096774 0.0766818700114025 0.22943548387096774 0.07867730900798175 0.22903225806451613 0.07896237172177879 0.22903225806451613 0.07981755986316989 0.22862903225806452 0.08010262257696693 0.22862903225806452 0.08124287343215507 0.22822580645161292 0.0815279361459521 0.22822580645161292 0.08523375142531357 0.22782258064516128 0.08551881413911061 0.22782258064516128 0.08608893956670467 0.22701612903225807 0.08665906499429875 0.21774193548387097 0.08665906499429875 0.21733870967741936 0.08694412770809579 0.2129032258064516 0.08694412770809579 0.2125 0.08665906499429875 0.20967741935483872 0.08665906499429875 0.20927419354838708 0.08637400228050171 0.20887096774193548 0.08637400228050171 0.20846774193548387 0.08608893956670467 0.18346774193548387 0.08608893956670467 0.18306451612903227 0.08580387685290763 0.18266129032258063 0.08580387685290763 0.18225806451612903 0.08608893956670467 0.18024193548387096 0.08608893956670467 0.17983870967741936 0.08580387685290763 0.1620967741935484 0.08580387685290763 0.1620967741935484 0.08694412770809579 0.16370967741935483 0.08694412770809579 0.16411290322580646 0.08665906499429875 0.17379032258064517 0.08665906499429875 0.17419354838709677 0.08637400228050171 0.175 0.08637400228050171 0.17540322580645162 0.08665906499429875 0.20443548387096774 0.08665906499429875 0.20483870967741935 0.08694412770809579 0.20524193548387096 0.08694412770809579 0.2056451612903226 0.08722919042189281 0.2056451612903226 0.08808437856328392 0.22862903225806452 0.08808437856328392 0.22862903225806452 0.08694412770809579 0.22943548387096774 0.08637400228050171 0.2310483870967742 0.08637400228050171 0.2310483870967742 0.07041049030786774 0.22983870967741934 0.07041049030786774 0.22943548387096774 0.0701254275940707 0.22943548387096774 0.06898517673888255 0.22983870967741934 0.06870011402508552 0.2310483870967742 0.06870011402508552 0.2310483870967742 0.06584948688711517 0.22983870967741934 0.06584948688711517 0.22943548387096774 0.06556442417331813 0.22943548387096774 0.0652793614595211 0.22983870967741934 0.06499429874572406 0.2310483870967742 0.06499429874572406 0.2310483870967742 0.033637400228050174
1 0.1620967741935484 0.033637400228050174 0.1620967741935484 0.03848346636259977 0.16411290322580646 0.03848346636259977 0.16451612903225807 0.03819840364880274 0.1685483870967742 0.03819840364880274 0.1689516129032258 0.03848346636259977 0.1786290322580645 0.03848346636259977 0.17903225806451614 0.03819840364880274 0.1814516129032258 0.03819840364880274 0.18185483870967742 0.03848346636259977 0.18951612903225806 0.03848346636259977 0.18991935483870967 0.03819840364880274 0.20120967741935483 0.03819840364880274 0.20161290322580644 0.03848346636259977 0.22661290322580646 0.03848346636259977 0.22701612903225807 0.03819840364880274 0.22943548387096774 0.03819840364880274 0.22943548387096774 0.033637400228050174
1 0.1620967741935484 0.03819840364880274 0.1620967741935484 0.04475484606613455 0.17620967741935484 0.04475484606613455 0.17620967741935484 0.04389965792474344 0.17661290322580644 0.043614595210946405 0.17983870967741936 0.043614595210946405 0.18024193548387096 0.04389965792474344 0.18024193548387096 0.04475484606613455 0.1810483870967742 0.04475484606613455 0.1810483870967742 0.04389965792474344 0.1814516129032258 0.043614595210946405 0.18669354838709679 0.043614595210946405 0.1870967741935484 0.043329532497149374 0.18911290322580646 0.043329532497149374 0.18951612903225806 0.043614595210946405 0.19233870967741937 0.043614595210946405 0.19274193548387097 0.04389965792474344 0.19274193548387097 0.04475484606613455 0.22943548387096774 0.04475484606613455 0.22943548387096774 0.03819840364880274
1 0.1620967741935484 0.04389965792474344 0.1620967741935484 0.04903078677309008 0.1774193548387097 0.04903078677309008 0.1778225806451613 0.04874572405929305 0.2 0.04874572405929305 0.20040322580645162 0.04903078677309008 0.20806451612903226 0.04903078677309008 0.20846774193548387 0.04874572405929305 0.22298387096774194 0.04874572405929305 0.22338709677419355 0.04903078677309008 0.22701612903225807 0.04903078677309008 0.22782258064516128 0.04846066134549601 0.22943548387096774 0.04846066134549601 0.22943548387096774 0.04389965792474344
1 0.1620967741935484 0.04846066134549601 0.1620967741935484 0.05501710376282782 0.22943548387096774 0.05501710376282782 0.22943548387096774 0.049885974914481185 0.22822580645161292 0.049885974914481185 0.22741935483870968 0.049315849486887116 0.22741935483870968 0.04846066134549601
1 0.1620967741935484 0.05416191562143672 0.1620967741935484 0.059293044469783354 0.18185483870967742 0.059293044469783354 0.18225806451612903 0.059578107183580385 0.18951612903225806 0.059578107183580385 0.18991935483870967 0.059293044469783354 0.22741935483870968 0.059293044469783354 0.22782258064516128 0.059007981755986316 0.22943548387096774 0.059007981755986316 0.22943548387096774 0.05416191562143672
1 0.1620967741935484 0.05872291904218928 0.1620967741935484 0.06499429874572406 0.16370967741935483 0.06499429874572406 0.16411290322580646 0.06470923603192702 0.175 0.06470923603192702 0.17540322580645162 0.06442417331812998 0.2161290322580645 0.06442417331812998 0.21653225806451612 0.06470923603192702 0.22661290322580646 0.06470923603192702 0.22701612903225807 0.06442417331812998 0.22741935483870968 0.06442417331812998 0.22741935483870968 0.06413911060433296 0.22782258064516128 0.06385404789053592 0.22782258064516128 0.06356898517673888 0.22822580645161292 0.06328392246294184 0.22943548387096774 0.06328392246294184 0.22943548387096774 0.06043329532497149 0.22822580645161292 0.06043329532497149 0.22701612903225807 0.059578107183580385 0.22701612903225807 0.05872291904218928
1 0.1620967741935484 0.06442417331812998 0.1620967741935484 0.0709806157354618 0.18508064516129033 0.0709806157354618 0.18508064516129033 0.0701254275940707 0.18548387096774194 0.06984036488027366 0.21129032258064517 0.06984036488027366 0.21169354838709678 0.0701254275940707 0.21169354838709678 0.0709806157354618 0.22701612903225807 0.0709806157354618 0.22701612903225807 0.06984036488027366 0.22741935483870968 0.06955530216647662 0.22741935483870968 0.06841505131128849 0.22782258064516128 0.06812998859749145 0.22782258064516128 0.06442417331812998
1 0.1620967741935484 0.0701254275940707 0.1620967741935484 0.0749714937286203 0.16532258064516128 0.0749714937286203 0.16572580645161292 0.07468643101482327 0.17459677419354838 0.07468643101482327 0.175 0.0749714937286203 0.17983870967741936 0.0749714937286203 0.18024193548387096 0.07468643101482327 0.1971774193548387 0.07468643101482327 0.1975806451612903 0.0749714937286203 0.22379032258064516 0.0749714937286203 0.22419354838709676 0.07525655644241733 0.22661290322580646 0.07525655644241733 0.22701612903225807 0.0749714937286203 0.22741935483870968 0.0749714937286203 0.22782258064516128 0.07468643101482327 0.22943548387096774 0.07468643101482327 0.22943548387096774 0.07183580387685291 0.22822580645161292 0.07183580387685291 0.22782258064516128 0.07155074116305588 0.22782258064516128 0.0701254275940707
1 0.1620967741935484 0.07468643101482327 0.1620967741935484 0.07981755986316989 0.16491935483870968 0.07981755986316989 0.16532258064516128 0.08010262257696693 0.1685483870967742 0.08010262257696693 0.1689516129032258 0.08038768529076397 0.1689516129032258 0.08124287343215507 0.22741935483870968 0.08124287343215507 0.22741935483870968 0.08038768529076397 0.22822580645161292 0.07981755986316989 0.22943548387096774 0.07981755986316989 0.22943548387096774 0.07468643101482327
1 0.16370967741935483 0.08038768529076397 0.16370967741935483 0.08580387685290763 0.20483870967741935 0.08580387685290763 0.20524193548387096 0.08608893956670467 0.20524193548387096 0.08694412770809579 0.22741935483870968 0.08694412770809579 0.22741935483870968 0.08608893956670467 0.22782258064516128 0.08580387685290763 0.22782258064516128 0.08551881413911061 0.22822580645161292 0.08523375142531357 0.22943548387096774 0.08523375142531357 0.22943548387096774 0.08409350057012542 0.22822580645161292 0.08409350057012542 0.22782258064516128 0.0838084378563284 0.22782258064516128 0.08352337514253136 0.22822580645161292 0.08323831242873432 0.22943548387096774 0.08323831242873432 0.22943548387096774 0.08038768529076397

EDIT: I think this is the answer.

1 reply

glenn-jocher Oct 3, 2024
Maintainer

@johnlockejrr it looks like you're on the right track with sorting the polygons by their topmost y-coordinate. To ensure they are sorted top-down, verify that the sorting key accurately reflects the topmost point of each polygon. If you need further assistance, feel free to ask!

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Ultralytics

Convert a YOLOv8 seg model detection result to YOLOv8 label #16658

{{title}}

{{editor}}'s edit

{{editor}}'s edit

Replies: 3 comments 1 reply

{{title}}

{{title}}

{{title}}

{{editor}}'s edit

{{editor}}'s edit

{{title}}

Select a reply

Ultralytics

Convert a YOLOv8 seg model detection result to YOLOv8 label #16658

johnlockejrr Oct 3, 2024

Replies: 3 comments · 1 reply

UltralyticsAssistant Oct 3, 2024 Maintainer

Upgrade

Environments

glenn-jocher Oct 3, 2024 Maintainer

johnlockejrr Oct 3, 2024 Author

glenn-jocher Oct 3, 2024 Maintainer

johnlockejrr
Oct 3, 2024

Replies: 3 comments 1 reply

UltralyticsAssistant
Oct 3, 2024
Maintainer

glenn-jocher
Oct 3, 2024
Maintainer

johnlockejrr
Oct 3, 2024
Author

glenn-jocher Oct 3, 2024
Maintainer