Axe-08 commited on
Commit
5d6e271
·
verified ·
1 Parent(s): 7e53a4e

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +55 -0
README.md ADDED
@@ -0,0 +1,55 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # IndoLepAtlas — Data Card
2
+
3
+ ## Dataset Summary
4
+
5
+ IndoLepAtlas is a computer vision dataset of Indian butterflies, moths, and their larval host plants. It captures the diversity of Indian Lepidoptera across ~1,094 species with images sourced from ifoundbutterflies.org.
6
+
7
+ **Inspired by:** iNaturalist
8
+ **Domain:** Indian Wildlife & Biodiversity (CSE3292 §3.6)
9
+
10
+ ## Source
11
+
12
+ - **Website:** [ifoundbutterflies.org](https://www.ifoundbutterflies.org/)
13
+ - **License:** Images are publicly available; photographer credits preserved
14
+ - **Collection method:** Automated scraping with respectful rate limiting
15
+
16
+ ## Dataset Composition
17
+
18
+ ### Butterflies (~967 species)
19
+ - Adult/Unknown life stage photographs
20
+ - Early stage (caterpillar, pupa) photographs
21
+ - Per-image metadata: species, common name, sex, location, date, credit
22
+ - Taxonomic hierarchy: Order → Superfamily → Family → Subfamily → Tribe → Genus → Species
23
+
24
+ ### Host Plants (~127 species)
25
+ - Hero images and gallery photographs
26
+ - Per-image metadata: species, family, location, date, credit
27
+ - Cross-referenced with butterfly species they host
28
+
29
+ ## Geographic Coverage
30
+
31
+ Images span all major Indian states with concentration in:
32
+ - Kerala, Karnataka, Maharashtra (Western Ghats biodiversity hotspot)
33
+ - Northeast India (Assam, Arunachal Pradesh, Meghalaya)
34
+ - Andaman & Nicobar Islands
35
+
36
+ ## Annotation Process
37
+
38
+ 1. **Auto-trimming:** Image overlays (text labels) removed via fixed-percentage cropping
39
+ 2. **Auto-detection:** Bounding boxes generated using Grounding DINO (zero-shot object detection)
40
+ 3. **Classification:** Species labels from source website taxonomy
41
+ 4. **Metadata enrichment:** OCR extraction of overlay text (location, date, credit, media code)
42
+
43
+ ## Known Limitations
44
+
45
+ - Some species have very few images (long-tailed distribution)
46
+ - OCR may fail on low-quality or unusually formatted overlays → null fields
47
+ - Bounding boxes are auto-generated, not manually verified
48
+ - Geographic bias toward well-surveyed regions (Western Ghats, Northeast)
49
+
50
+ ## Ethical Considerations
51
+
52
+ - No human subjects or personal information
53
+ - Source is a public biodiversity documentation platform
54
+ - Photographer attribution maintained via OCR extraction
55
+ - Dataset does not include sensitive or endangered species location details