object detection data coordinate convert, {polygon, center) -> (left, top, right, bottom)

 

I recently download some object detection dataset from Roboflow. 

But the coordinate is strange, some coordinate is polygon, some is cenxter x,y and width, height.

My favourite coordinate is left, top, right, bottom.

So this code figure out type of coordinate and covert it to [left, top, right, bottom]

Thank you.

.

import cv2

import numpy as np

import os

from tqdm import tqdm

def convert_and_write_coordinates(txt_path, rect_coords, poly_coords, w, h):

"""Converts coordinates to (label, left, top, right, bottom) and writes them back in a normalized format."""

with open(txt_path, 'w') as file:

# Write rectangle coordinates in normalized (label, left, top, right, bottom) format

for coord_set in rect_coords:

center_x, center_y, width, height = coord_set

left = (center_x - width / 2) * w

top = (center_y - height / 2) * h

right = (center_x + width / 2) * w

bottom = (center_y + height / 2) * h

# Normalize coordinates

left_norm = left / w

top_norm = top / h

right_norm = right / w

bottom_norm = bottom / h

# Write the rectangle coordinates in normalized format

file.write(f"0 {left_norm} {right_norm} {top_norm} {bottom_norm}\n")

# Write polygon coordinates in normalized (label, left, top, right, bottom) format

for points in poly_coords:

# Convert polygon points to bounding box

x_coords = [x * w for x, y in points]

y_coords = [y * h for x, y in points]

left = min(x_coords)

top = min(y_coords)

right = max(x_coords)

bottom = max(y_coords)

# Normalize coordinates

left_norm = left / w

top_norm = top / h

right_norm = right / w

bottom_norm = bottom / h

# Write the polygon coordinates in normalized bounding box format

file.write(f"0 {left_norm} {right_norm} {top_norm} {bottom_norm}\n")

def draw_annotations(image_path, txt_path):

# Load the image

image = cv2.imread(image_path)

if image is None:

print(f"Failed to load image: {image_path}")

return False

# Get image dimensions

h, w, _ = image.shape

# Read the annotation data from the corresponding text file

rect_coords = [] # To store rectangle coordinates

poly_coords = [] # To store polygon coordinates

with open(txt_path, 'r') as file:

lines = file.readlines()

for line in lines:

parts = line.strip().split()

label = int(parts[0]) # Extract the label index

coords = [float(p) for p in parts[1:]]

if len(coords) == 4: # Assuming it's (center_x, center_y, width, height)

rect_coords.append((coords[0], coords[1], coords[2], coords[3]))

elif len(coords) % 2 == 0: # Assuming pairs of coordinates for a polygon

points = [(coords[i], coords[i + 1]) for i in range(0, len(coords), 2)]

poly_coords.append(points)

# Write the new coordinates to the text file

convert_and_write_coordinates(txt_path, rect_coords, poly_coords, w, h)

return True

def process_images_in_folder(folder_path):

image_files = [f for f in os.listdir(folder_path) if f.lower().endswith(('.jpg', '.png'))]

for filename in tqdm(image_files, desc="Processing images"):

image_path = os.path.join(folder_path, filename)

txt_path = os.path.join(folder_path, os.path.splitext(filename)[0] + '.txt')

if os.path.isfile(txt_path):

draw_annotations(image_path, txt_path)

else:

print(f"No corresponding text file found for image: {filename}")

def main():

# Specify the directory containing the images and text files

folder_path = './val_indonesia_roboflow_2024_08_03/'

# Process the images in the folder

process_images_in_folder(folder_path)

if __name__ == "__main__":

main()

..

bonus code 

Display image + annotation rect box

.

import cv2

import os

from tqdm import tqdm

def display_image_with_annotations(image_path, txt_path):

"""Display the image with rectangle annotations."""

# Load the image

image = cv2.imread(image_path)

if image is None:

print(f"Failed to load image: {image_path}")

return False

# Get image dimensions

h, w, _ = image.shape

# Read the annotation data from the text file

with open(txt_path, 'r') as file:

lines = file.readlines()

for line in lines:

parts = line.strip().split()

if len(parts) == 5: # Check for valid rectangle data

# Extract label and normalized coordinates

label = int(parts[0]) # The label is currently unused, always 0 in your case

left_norm, right_norm, top_norm, bottom_norm = map(float, parts[1:])

# Convert normalized coordinates to absolute pixel coordinates

left = int(left_norm * w)

right = int(right_norm * w)

top = int(top_norm * h)

bottom = int(bottom_norm * h)

# Draw rectangle on the image

cv2.rectangle(image, (left, top), (right, bottom), (255, 0, 0), 2)

# Display the image with annotations

cv2.imshow('Annotated Image', image)

key = cv2.waitKey(0)

cv2.destroyAllWindows()

# Return True if 'q' was pressed, otherwise False

return key == ord('q')

def process_images_in_folder(folder_path):

image_files = [f for f in os.listdir(folder_path) if f.lower().endswith(('.jpg', '.png'))]

for filename in tqdm(image_files, desc="Processing images"):

image_path = os.path.join(folder_path, filename)

txt_path = os.path.join(folder_path, os.path.splitext(filename)[0] + '.txt')

if os.path.isfile(txt_path):

if display_image_with_annotations(image_path, txt_path):

print("Exiting image display loop.")

break

else:

print(f"No corresponding text file found for image: {filename}")

def main():

# Specify the directory containing the images and text files

folder_path = './train_indonesia_roboflow_2024_08_03/'

# Process the images in the folder

process_images_in_folder(folder_path)

if __name__ == "__main__":

main()

..

marearts

🙇🏻‍♂️