Hugging Face's logo

Datasets:
iscc
/
iscc-flickr30k

Tasks:
Image Feature Extraction
Zero-Shot Image Classification
Modalities:
Image
Text
Formats:
parquet
Languages:
English
Size:
10K - 100K
Tags:
iscc
content-identification
similarity-search
deduplication
image
iso-24138
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
Community
1
iscc-flickr30k / README.md
titusz's picture
titusz
Update dataset card
f760e28 verified
preview code
|
raw
history blame contribute delete
8.73 kB
metadata 
annotations_creators:
  - machine-generated
language_creators:
  - found
language:
  - en
license: other
multilinguality:
  - monolingual
size_categories:
  - 10K<n<100K
source_datasets:
  - lmms-lab/flickr30k
task_categories:
  - image-feature-extraction
  - zero-shot-image-classification
tags:
  - iscc
  - content-identification
  - similarity-search
  - deduplication
  - image
  - iso-24138
  - flickr30k
pretty_name: ISCC Codes for Flickr30k
dataset_info:
  features:
    - name: thumbnail
      dtype: image
    - name: source_dataset
      dtype: string
    - name: source_row_id
      dtype: string
    - name: filename
      dtype: string
    - name: caption
      dtype: string
    - name: iscc
      dtype: string
    - name: iscc_meta
      dtype: string
    - name: iscc_semantic
      dtype: string
    - name: iscc_content
      dtype: string
    - name: iscc_data
      dtype: string
    - name: iscc_instance
      dtype: string
    - name: width
      dtype: int32
    - name: height
      dtype: int32
    - name: filesize
      dtype: int64
    - name: image_path
      dtype: string

ISCC Codes for Flickr30k

The Flickr30k dataset enriched with full 256-bit ISCC (International Standard Content Code) codes for content identification, similarity search, and deduplication research.

What is ISCC?

The International Standard Content Code (ISO 24138:2024) is a content-derived identifier for digital media assets. Unlike traditional identifiers that are assigned arbitrarily, ISCC codes are generated algorithmically from the content itself, enabling:

  • Content Identification: Identify content regardless of format or location
  • Similarity Search: Find visually or semantically similar images
  • Deduplication: Detect exact and near-duplicate content
  • Provenance Tracking: Link derived works to their sources

ISCC Units

Each record contains five 256-bit ISCC-UNITs that capture different aspects of the content:

Unit Field Description
Meta-Code iscc_meta Similarity based on embedded metadata (filename, title)
Semantic-Code iscc_semantic AI-based visual semantic similarity (what the image depicts)
Content-Code iscc_content Perceptual image similarity (visual appearance)
Data-Code iscc_data Raw binary data similarity (file structure)
Instance-Code iscc_instance Cryptographic hash for exact matching (like SHA-256)

The iscc field contains the composite ISCC-CODE combining all units.

Dataset Structure

Data Fields

Field Type Description
thumbnail image WebP thumbnail (128x128 max) for preview
source_dataset string Source HuggingFace dataset path (lmms-lab/flickr30k)
source_row_id string Original row identifier in source dataset
filename string Original filename (e.g., 1000092795.jpg)
caption string First of 5 human-written captions from Flickr30k
iscc string Full composite ISCC-CODE
iscc_meta string 256-bit Meta-Code
iscc_semantic string 256-bit Semantic-Code
iscc_content string 256-bit Content-Code
iscc_data string 256-bit Data-Code
iscc_instance string 256-bit Instance-Code
width int Original image width in pixels
height int Original image height in pixels
filesize int File size in bytes
image_path string Relative path to cached source image

Data Splits

Split Samples
train 31,783

Usage

Loading the Dataset 

from datasets import load_dataset

ds = load_dataset("iscc/iscc-flickr30k")

Viewing a Sample 

sample = ds["train"][0]
print(f"ISCC: {sample['iscc']}")
print(f"Dimensions: {sample['width']}x{sample['height']}")

Similarity Search Example 

import iscc_core as ic

# Get two ISCC codes to compare
code1 = ds["train"][0]["iscc_content"]
code2 = ds["train"][1]["iscc_content"]

# Calculate hamming distance (0 = identical, 256 = maximally different)
distance = ic.iscc_distance(code1, code2)
print(f"Hamming distance: {distance}")

# Convert to similarity percentage
similarity = 1 - (distance / 256)
print(f"Similarity: {similarity:.1%}")

Finding Near-Duplicates 

import iscc_core as ic

# Threshold for near-duplicates (adjust based on use case)
THRESHOLD = 32  # ~87.5% similarity

reference = ds["train"][0]["iscc_content"]

for i, row in enumerate(ds["train"]):
    distance = ic.iscc_distance(reference, row["iscc_content"])
    if distance <= THRESHOLD and i > 0:
        print(f"Near-duplicate found: row {i}, distance={distance}")

Semantic Similarity Search 

import iscc_core as ic

# Find semantically similar images (same subject/concept)
reference = ds["train"][0]["iscc_semantic"]

similar = []
for i, row in enumerate(ds["train"]):
    distance = ic.iscc_distance(reference, row["iscc_semantic"])
    if distance <= 64:  # ~75% semantic similarity
        similar.append((i, distance))

# Sort by similarity
for idx, dist in sorted(similar, key=lambda x: x[1])[:5]:
    print(f"Row {idx} (distance={dist})")

Source Data

This dataset was derived from lmms-lab/flickr30k.

Original Flickr30k Dataset

The Flickr30k dataset contains 31,783 images collected from Flickr, each with 5 human-written captions. It is widely used for image captioning and visual-semantic research. License: The original Flickr30k images are subject to Flickr's Terms of Service. This derivative dataset contains only ISCC codes and small 128px thumbnails for visual verification of matches, not the original high-resolution images. Users should refer to the original dataset for licensing details.

Processing

ISCC codes were generated using:

  • iscc-sdk - High-level ISCC generation
  • iscc-sci - Semantic image codes (experimental)

All processing was performed on original resolution images. WebP thumbnails (128x128 max, quality 80) were generated separately for dataset preview purposes.

Considerations

Intended Use

  • Content identification and matching research
  • Image similarity search algorithm development
  • Deduplication system benchmarking
  • Visual-semantic retrieval experiments
  • ISCC-based indexing research

Limitations

  • Semantic codes are generated using experimental AI models and may not capture all semantic nuances
  • ISCC codes are sensitive to significant image modifications (heavy cropping, overlays, filters)
  • Thumbnails are for preview only; use the source dataset for full-resolution images

Privacy

This dataset contains ISCC codes and thumbnails derived from the source images. Refer to the original dataset documentation for privacy considerations.

Citation

If you use this dataset, please cite both this dataset and the original source:

This Dataset:

@dataset{iscc_flickr30k,
  title = {{ISCC Codes for Flickr30k}},
  author = {{ISCC Foundation}},
  year = {{2026}},
  publisher = {{Hugging Face}},
  url = {{https://huggingface.co/datasets/iscc/iscc-flickr30k}}
}

Original Flickr30k:

@article{young2014image,
  title = {{From image descriptions to visual denotations: New similarity metrics
  for semantic inference over event descriptions}},
  author = {{Young, Peter and Lai, Alice and Hodosh, Micah and Hockenmaier, Julia}},
  journal = {{Transactions of the Association for Computational Linguistics}},
  volume = {{2}},
  pages = {{67--78}},
  year = {{2014}},
  publisher = {{MIT Press}}
}

ISCC Standard:

@misc{iso24138,
  title = {{ISO 24138:2024 Information and documentation -- International Standard Content Code (ISCC)}},
  author = {{International Organization for Standardization}},
  year = {{2024}},
  url = {{https://www.iso.org/standard/77899.html}}
}

Additional Resources

Contact