dataproc5/metrics-danbooru2025-alltime-tag-counts

Dataset Overview

tag_count provides aggregated tag usage statistics from the Danbooru2025 dataset. Each entry corresponds to a specific tag's usage count in all time.

import unibox as ub

df = ub.loads("hf://dataproc5/metrics-danbooru2025-monthly-tag-counts").to_pandas()
alltime_tag_counts = df.groupby(["tag_string", "tag_type"], as_index=False)["tag_count"].sum()
alltime_tag_counts = alltime_tag_counts.sort_values("tag_count", ascending=False)
ub.saves(alltime_tag_counts, "hf://dataproc5/metrics-danbooru2025-alltime-tag-counts", private=False)

Usage

Some example use cases of this metrics includes:

• finding out the top-occuring character / artist tags for targeted finetunes
• creating tag-balanced datasets
• use as a weighted random tags generator

Columns

• tag_string: The text of the tag (e.g., "landscape").

• tag_count: The total occurrences of the tag in the entire Danbooru dataset

• tag_type: The category of the tag:

  • "artist": Artist names.
  • "character": Character names.
  • "copyright": Copyrighted works or IPs.
  • "general": General descriptive tags.
  • "meta": Meta information tags.

Source Data

• Derived from Danbooru2025 image metadata.
• Tags are extracted from columns: tag_string_artist, tag_string_character, tag_string_copyright, tag_string_general, and tag_string_meta.

Visualization

Danbooru tag counts are highly unbalanced. using df["tag_count"].hist() barely gets anything. Here's a log-scaled vis for reference:

import matplotlib.pyplot as plt

# Plot histogram with log scale on the y-axis
plt.figure(figsize=(10, 6))
plt.hist(alltime_tag_counts["tag_count"], bins=100, log=True, edgecolor='black')
plt.title("Histogram of All-Time Tag Counts (Log Scale)")
plt.xlabel("Tag Count")
plt.ylabel("Log Frequency")
plt.grid(True)
plt.show()