htr_demo / src /htr_pipeline /utils /order_of_object.py
Gabriel's picture
added new dataset
417b347
raw
history blame contribute delete
No virus
3.59 kB
import numpy as np
import pandas as pd
class OrderObject:
def __init__(self):
pass
def order_lines(self, line_image, line_spacing_factor=0.5):
bounding_boxes = line_image.pred_instances.bboxes.tolist()
center_points = [(box[1] + box[3]) / 2 for box in bounding_boxes]
horizontal_positions = [(box[0] + box[2]) / 2 for box in bounding_boxes]
# Calculate the threshold distance
threshold_distance = self._calculate_threshold_distance(bounding_boxes, line_spacing_factor)
# Sort the indices based on vertical center points and horizontal positions
indices = list(range(len(bounding_boxes)))
indices.sort(
key=lambda i: (
center_points[i] // threshold_distance,
horizontal_positions[i],
)
)
# Order text lines
return indices
def _calculate_threshold_distance(self, bounding_boxes, line_spacing_factor=0.5):
# Calculate the average height of the text lines
total_height = sum(box[3] - box[1] for box in bounding_boxes)
average_height = total_height / len(bounding_boxes)
# Calculate the threshold distance, Set a factor for the threshold distance (adjust as needed)
threshold_distance = average_height * line_spacing_factor
# Return the threshold distance
return threshold_distance
def order_regions_marginalia(self, region_image, margin_ratio=0.2, histogram_bins=50, histogram_dip_ratio=0.5):
bounding_boxes = region_image.pred_instances.bboxes.tolist()
img_width = region_image.metainfo["ori_shape"][1]
regions = [[i, x[0], x[1], x[0] + x[2], x[1] + x[3]] for i, x in enumerate(bounding_boxes)]
# Create a pandas DataFrame from the regions
df = pd.DataFrame(regions, columns=["region_id", "x_min", "y_min", "x_max", "y_max"])
# Calculate the centroids of the bounding boxes
df["centroid_x"] = (df["x_min"] + df["x_max"]) / 2
df["centroid_y"] = (df["y_min"] + df["y_max"]) / 2
# Calculate a histogram of the x-coordinates of the centroids
histogram, bin_edges = np.histogram(df["centroid_x"], bins=histogram_bins)
# Determine if there's a significant dip in the histogram, which would suggest a two-page layout
is_two_pages = np.min(histogram) < np.max(histogram) * histogram_dip_ratio
if is_two_pages:
# Determine which page each region is on
page_width = int(img_width / 2)
df["page"] = (df["centroid_x"] > page_width).astype(int)
# Determine if the region is in the margin
margin_width = page_width * margin_ratio
df["is_margin"] = ((df["page"] == 0) & (df["centroid_x"] < margin_width)) | (
(df["page"] == 1) & (df["centroid_x"] > img_width - margin_width)
)
else:
df["page"] = 0
df["is_margin"] = (df["centroid_x"] < img_width * margin_ratio) | (
df["centroid_x"] > img_width - page_width * margin_ratio
)
# Define a custom sorting function
sort_regions = lambda row: (
row["page"],
row["is_margin"],
row["centroid_y"],
row["centroid_x"],
)
# Sort the DataFrame using the custom function
df["sort_key"] = df.apply(sort_regions, axis=1)
df = df.sort_values("sort_key")
# Return the ordered regions
return df["region_id"].tolist()
if __name__ == "__main__":
pass