v1.1.0-real-dataset

v1.1.0-real-dataset

Real-photo iNaturalist-trained Snake Detector demo model used by the live Hugging Face Space.

Live app: https://mmaitland-snake-detector-demo.hf.space
Space repo: https://huggingface.co/spaces/mmaitland/snake-detector-demo

Included Assets

• model.keras real-photo Keras model
• deployment_config.json live inference config
• run_config.json training/run config
• metrics_test_t076.json held-out test metrics at threshold 0.76
• threshold_sweep_val.json validation threshold sweep
• confusion_matrix_test_t076.png
• training_curves.png
• sample_predictions_test_t076.png
• sample_predictions_test_t076.json
• SHA256SUMS.txt checksum manifest

Held-Out Test Summary

• Test split: 1,283 iNaturalist-sourced images (255 snake, 1,028 no-snake)
• Accuracy: 0.9026
• Macro F1: 0.8358
• Snake precision: 0.8095
• Snake recall: 0.6667
• Snake F1: 0.7312

Claim Boundary

This is a bounded binary snake vs no_snake demo. It is not species identification, field-ready wildlife handling guidance, or a safety device. Raw third-party image files are not included in this release.

Full release package documentation: https://github.com/mmaitland300/Snake-detector/blob/main/docs/releases/v1.1.0-real-dataset.md