|
--- |
|
license: agpl-3.0 |
|
pipeline_tag: object-detection |
|
tags: |
|
- ultralytics |
|
- tracking |
|
- instance-segmentation |
|
- image-classification |
|
- pose-estimation |
|
- obb |
|
- object-detection |
|
- yolo |
|
- yolov8 |
|
- license_plate |
|
- Iran |
|
- veichle_lisence_plate |
|
- onnx |
|
--- |
|
|
|
|
|
## Model Overview |
|
This model is an ONNX version of a [YOLOv8 medium fine tuned](https://huggingface.co/shalchianmh/Iran_license_plate_detection_YOLOv8m) on Iran veichle plate dataset. YOLOv8 is designed to efficiently detect objects in images by generating bounding boxes around objects of interest and predicting their associated class probabilities. |
|
|
|
|
|
## How to Use |
|
|
|
### Inference Using ONNX Runtime |
|
```python |
|
import onnxruntime as rt |
|
sess = rt.InferenceSession("path_to_model.onnx") |
|
|
|
# View model input and output details |
|
input_name = sess.get_inputs()[0].name |
|
print("Input name:", input_name) |
|
input_shape = sess.get_inputs()[0].shape |
|
print("Input shape:", input_shape) |
|
input_type = sess.get_inputs()[0].type |
|
print("Input type:", input_type) |
|
|
|
output_name = sess.get_outputs()[0].name |
|
print("Output name:", output_name) |
|
output_shape = sess.get_outputs()[0].shape |
|
print("Output shape:", output_shape) |
|
output_type = sess.get_outputs()[0].type |
|
print("Output type:", output_type) |
|
``` |
|
|
|
### Pre-processing |
|
1. **Load Image**: Load the image using `cv2.imread()`. |
|
2. **Resize**: Resize the input image to a 224x224 resolution. |
|
3. **Normalize**: Scale pixel values to the range [0, 1] by dividing by 255. |
|
4. **Transpose**: Change the image array to channel-first format `(C, H, W)`. |
|
5. **Convert**: Convert the image to a float32 NumPy array and add a batch dimension. |
|
|
|
```python |
|
import cv2 |
|
import numpy as np |
|
|
|
# Pre-process the input image |
|
image_path = "/path_to_image.png" |
|
input_image = cv2.imread(image_path) |
|
resized_image = cv2.resize(input_image, (224, 224)) |
|
scaled_image = resized_image / 255.0 |
|
transposed_image = scaled_image.transpose((2, 0, 1)) |
|
prep_image = np.array(transposed_image, dtype='float32')[None, :] |
|
``` |
|
|
|
### Inference |
|
After pre-processing, run the model on the prepared image. |
|
|
|
```python |
|
# Run inference |
|
output_probabilities = sess.run([output_name], {input_name: prep_image}) |
|
``` |
|
|
|
### Post-processing |
|
To extract bounding box details: |
|
- **Identify** the most probable bounding box. |
|
- **Extract** the coordinates (x, y, width, height) and the associated probability. |
|
|
|
```python |
|
# Extract bounding box information |
|
most_prob_idx = output_probabilities[0][0][4].argmax() |
|
x, y, width, height, prob = output_probabilities[0][0][:, most_prob_idx] |
|
print(x, y, width, height, prob) |
|
``` |
|
|
