EcoID: An Offline Plant Identifier That Gets You Outside

Описание и применение

EcoID – автономный инструмент распознавания растений, работающий без подключения к интернету, с локальным хранением данных и ручной проверкой результатов.

Задача

Field researchers and citizen scientists need a privacy‑preserving, offline plant identification tool that can run on a laptop or phone and store observations locally.

Результат

EcoID achieves 86.0% Top‑1 accuracy on a 50‑image evaluation set, 94.0% Top‑3 accuracy, and 12.4 ms total inference latency on a Windows laptop, while keeping all data and model weights on the device.

Рабочий процесс

• Capture or upload a photo
• Run local inference with ONNX Runtime
• Review ranked identification candidates
• Verify, reject, or mark uncertain
• Save the field observation
• Review observations in history and on the local map

Доказательства

EcoID is an offline‑first plant identification tool for field observations.

Источник

The current model supports five plant classes: Mango, Coconut, Banana, Papaya, Cassava.

Источник

The model was benchmarked on a Windows laptop using CPU inference: Total 12.4 ms, p95 33.8 ms per image.

Источник

On a separate 50‑image evaluation set (10 images per class), the model achieved: Top‑1 accuracy 86.0% (43/50).

Источник

The application stores everything locally: ~/.ecoid/ ├── ecoid.db ├── images/ ├── thumbs/ └── drafts/

Источник