README.md

metadata

title: Butterfly Classifier
emoji: 🦋
colorFrom: red
colorTo: blue
sdk: gradio
sdk_version: 5.44.1
app_file: app.py
pinned: false
short_description: Upload a photo to use a model for butterfly classification

🦋 Butterfly Classifier — Live Demo

Upload a butterfly photo and see the top-k predicted species with confidence scores.

Standalone app window: https://michaelmwb-butterfly-classifier.hf.space/

For access to dataset of butterfly images, go → here

---

How to use

1. Upload image (JPG/PNG).
2. Set Top-k (default: 5).
3. Click Predict → the app returns the top classes with confidences.

Device: The app selects the best device automatically (GPU if available on the Space, otherwise CPU). A manual CPU option is provided if needed.

---

For best results

• Single butterfly, centered, sharp focus
• Plain/clean background (avoid clutter)
• Show full wings; crop out large borders
• Good lighting; avoid harsh shadows or filters
• Image size around 800–1200 px on the short side

---

What this demo shows

• Top-k predictions with confidence scores
• Auto device selection (GPU/CPU) with a CPU fallback
• Clean, upload-only interface (webcam disabled for maximum compatibility)

---

Model & data (short card)

• Backbone: VGG16 (transfer learning)
• Classes: 75 butterfly species
• Training: fine-tuned classifier head; ImageNet mean/std normalization
• Validation performance: ~90% top-1 accuracy on held-out data
• Weights: hosted on the Hub → MichaelMwb/butterfly-vgg16

Note: Accuracy depends on photo quality and how close the image distribution is to the training set.

---

Privacy & disclaimer

• Privacy: Images are processed in memory and are not stored by this app.
• Disclaimer: This is a research demo. Predictions may be incorrect—use as suggestions, not definitive IDs.

---

Known limitations

• Webcam: intentionally disabled to avoid browser permission issues; upload works everywhere.
• Edge cases: unusual angles, heavy occlusion, extreme lighting, or multiple butterflies may reduce accuracy.

---