ONNX Quantized Model: crop_leaf_diseases_vit
This repository contains an ONNX quantized version of the pre-trained ViT model for crop leaf disease classification originally from wambugu71/crop_leaf_diseases_vit.
The original model was converted to the ONNX format and then quantized to INT8 for improved efficiency and reduced size, making it suitable for deployment in resource-constrained environments, such as mobile devices.
Model Details
- Original Model:
wambugu71/crop_leaf_diseases_vit - Format: ONNX (Quantized to INT8)
- Task: Image Classification
- Classes: (List the classes if available from the original model card or documentation, otherwise mention they are from the original model)
Usage
This model can be loaded and used for inference with the ONNX Runtime.
import onnxruntime as ort
import numpy as np
from PIL import Image
from transformers import AutoFeatureExtractor # Assuming feature_extractor is needed for preprocessing
# Load the quantized ONNX model
onnx_model_path = "crop_leaf_diseases_vit.quant.onnx" # or the correct quantized model file name
session = ort.InferenceSession(onnx_model_path)
# Get input and output names
input_name = session.get_inputs()[0].name
output_name = session.get_outputs()[0].name
# Example inference (assuming you have an image and a feature extractor)
# Replace with your actual image loading and preprocessing
image_path = "path/to/your/image.jpg"
image = Image.open(image_path).convert("RGB")
# Load the feature extractor used during conversion
# feature_extractor = AutoFeatureExtractor.from_pretrained("wambugu71/crop_leaf_diseases_vit")
# inputs = feature_extractor(images=image, return_tensors="np")
# pixel_values = inputs["pixel_values"]
# If not using a feature extractor, you might need to preprocess manually
# based on the original model's requirements (e.g., resize, normalize)
# For example, if the model expects 3x224x224 float32 input:
# resized_image = image.resize((224, 224))
# input_data = np.array(resized_image, dtype=np.float32).transpose(2, 0, 1) # HWC to CHW
# input_data = np.expand_dims(input_data, axis=0) # Add batch dimension
# pixel_values = input_data
# Run inference
outputs = session.run([output_name], {input_name: pixel_values})
logits = outputs[0]
# Post-processing (example for classification)
predicted_class_idx = np.argmax(logits, axis=1)[0]
print("Predicted class index:", predicted_class_idx)
# You would then map the predicted_class_idx to the actual class label
# using the label mapping from the original model if available.
Quantization Details
This model was quantized using dynamic quantization with ONNX Runtime.
Files
crop_leaf_diseases_vit.onnx: The original ONNX model (before quantization).crop_leaf_diseases_vit.quant.onnx: The quantized ONNX model (INT8) find colab link.