| ---
|
| language: en
|
| library_name: onnxruntime
|
| pipeline_tag: image-classification
|
| tags: [plant-disease, image-classification, onnx, onnxruntime, cpu-inference]
|
| ---
|
|
|
| # Cambium Plant Disease Detector
|
|
|
| CPU-optimized MobileNetV2 image classifier exported to ONNX and validated with ONNX Runtime.
|
|
|
| ## Model details
|
|
|
| - Runtime: ONNX Runtime
|
| - Execution provider: CPUExecutionProvider
|
| - Input: `pixel_values`, float32 NCHW tensor with shape `[1, 3, 224, 224]`
|
| - Preprocessing: RGB, resize to 257px, center crop to 224px, then ImageNet normalization.
|
| - Output: `logits`; map output indices using `idx_to_class.json`.
|
|
|
| ## Intended use
|
|
|
| Research and demonstration plant-disease image classification only. This is not a substitute for advice from a qualified agronomist or plant pathologist.
|
|
|
| ## Files
|
|
|
| - `model.onnx`: portable ONNX model.
|
| - `model_optimized.onnx`: ONNX Runtime graph-optimized CPU model.
|
| - `idx_to_class.json`: output-index mapping.
|
| - `metadata.json`: parity and latency validation results.
|
|
|
| ## Validation
|
|
|
| The export script compares PyTorch and ONNX logits and fails if maximum absolute error exceeds `1e-3`. It also benchmarks CPU inference and records mean and p95 latency in `metadata.json`.
|
|
|
