Datasets:

Horama
/

classif_background

license: other
task_categories:
  - image-classification
language:
  - fr
tags:
  - background
  - no-animal
  - classification
pretty_name: Animal Classification (Background)
size_categories:
  - 10K<n<100K

Background Dataset — `Horama/classif_background`

Dataset of images without animals, used as a background class.

This dataset is part of the merged dataset: 👉 Horama/classif_merged_187

Dataset Summary

Images with no animals present
Label: null

Intended Use

Train models to predict “no animal”
Reduce false positives
Improve classifier robustness

Data Source

Derived from COCO dataset

Splits

Predefined splits are preserved from the original dataset.

Data Processing

Same preprocessing pipeline as: 👉 Horama/classif_merged_187

Dataset Structure

Data is stored as zstd-compressed Parquet shards.

Images are stored as raw bytes and decoded on-the-fly.

Row format

{
    "image": <PIL Image>,  # stored as raw bytes
    "label": "null",
    "distance": str,
}

Biases & Limitations

May not reflect real-world negative samples Distribution differs from production pipeline

Ethical Considerations

Inherits constraints from COCO dataset

License

Follows COCO dataset licensing.

Background Dataset — Horama/classif_background