IndoLepAtlas / docs /README.md
AXE8's picture
Sync pipeline artifacts (metadata, annotations, splits, docs)
e42237f verified

IndoLepAtlas — Indian Lepidoptera & Host Plants Dataset

A large-scale computer vision dataset of Indian butterflies, moths, and their larval host plants. Sourced from ifoundbutterflies.org with public CC-licensed photographs.

Inspired by: iNaturalist | Domain: Indian Wildlife & Biodiversity

Dataset Overview

Butterflies Host Plants Total
Species ~967 ~127 ~1,094
Images ~2,000+ ~700+ ~2,700+
Source ifoundbutterflies.org ifoundbutterflies.org

Motivation

Existing CV datasets (ImageNet, iNaturalist) under-represent Indian biodiversity. India hosts ~1,500 butterfly species — many endemic and poorly documented in digital datasets. This dataset enables:

  • Species classification of Indian butterflies and host plants
  • Object detection with auto-generated bounding boxes
  • Ecological analysis of butterfly-plant host relationships
  • Diagnostic analysis using pretrained models

Directory Structure

IndoLepAtlas/
├── data/
│   ├── butterflies/
│   │   ├── raw/              # Original images (with overlays)
│   │   ├── images/           # Clean trimmed images
│   │   └── metadata.csv      # Enriched per-image metadata
│   └── plants/
│       ├── raw/
│       ├── images/
│       └── metadata.csv
├── annotations/
│   ├── butterflies/*.txt     # YOLO format
│   ├── plants/*.txt          # YOLO format
│   ├── annotations.json      # COCO format (combined)
│   └── classes.txt           # Species→ID mapping
├── splits/
│   ├── train.txt (70%)
│   ├── val.txt   (15%)
│   └── test.txt  (15%)
├── docs/
│   ├── class_definitions.json
│   ├── annotation_guide.md
│   ├── data_card.md
│   └── distribution_stats.md
├── README.md
└── LICENSE

Metadata Fields

Butterflies (metadata.csv)

image_id, filename, species, common_name, family, subfamily, genus, order, life_stage, sex, media_code, location, state, date, credit, source_url, source, split

Plants (metadata.csv)

image_id, filename, species, family, genus, image_type, media_code, location, state, date, credit, butterfly_hosts, source_url, source, split

Annotation Format

  • YOLO (per image): <class_id> <x_center> <y_center> <width> <height> (normalized)
  • COCO (annotations.json): standard format with images, categories, annotations
  • Classes: Species-level (~1,094 classes). See annotations/classes.txt

Pipeline Scripts

Script Purpose
pull_hf_data.py Download datasets from HuggingFace
process_images.py Trim overlay text bands
enrich_metadata.py OCR overlay extraction → metadata CSVs
generate_annotations.py Auto bounding boxes (Grounding DINO)
generate_splits.py Stratified train/val/test splits
generate_stats.py Distribution statistics
crawler_logged.py Butterfly species scraper
plant_crawler.py Host plant species scraper

All pipeline scripts are incremental — they skip already-processed data and support adding new species non-destructively.

Ethical Guidelines

  • All images are from publicly available sources (ifoundbutterflies.org)
  • Photographer credits are preserved in metadata via OCR extraction
  • No human faces or personal information in the dataset
  • Source URLs and attribution maintained for every image

Citation

@dataset{indolepatlas2026,
  title={IndoLepAtlas: Indian Lepidoptera and Host Plants Dataset},
  year={2026},
  source={ifoundbutterflies.org},
  license={CC BY-NC-SA 4.0}
}

License

This dataset is released under CC BY-NC-SA 4.0.