Skip to main content

Yolo App - Pipeline Predictions

Shenzhen, China

  1. Prepare your Images and get Data
  2. Train your Tensorflow Model
  3. Use your Model to do Predictions
  4. Use Tesseract to Read Number Plates
  5. Flask Web Application
  6. Yolo v5 - Data Prep

Model Predictions​

Loading the Model​

I can now load the trained model and try to run a prediction:

import numpy as np
import cv2
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras.preprocessing.image import load_img, img_to_array

# Load the trained model
model = tf.keras.models.load_model('../models/object_detection.h5')
print('[INFO] object detection model loaded')

Memory Error in Tensorflow: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:936] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero

Running the following command from my console to set each device to 0 got rid of the warning:

for a in /sys/bus/pci/devices/*; do echo 0 | sudo tee -a $a/numa_node; done

Load a Test Image​

I now need to pick an image the model was not trained on to check to models accuracy:

# Pick a test image that your model was not trained on
path = './test_images/index5.jpg'
image = load_img(path) # PIL object

# convert into array and get the normalized output
image = np.array(image,dtype=np.uint8) # 8 bit array (0,255)
image1 = load_img(path,target_size=(224,224))
image_arr_224 = img_to_array(image1)/255.0

# Get size of the orginal image
h,w,d = image.shape
print('Image height:',h)
print('Image width:',w)

# Show selected image
plt.figure(figsize=(10,8))
plt.imshow(image)
plt.show()

Predict Number Plate Coordinates​

# Check if dimensions are OK
image_arr_224.shape

# Add index number colume to shape
test_arr = image_arr_224.reshape(1,224,224,3)
test_arr.shape

# Use trained model to predict 
# number plate location
coords = model.predict(test_arr)
coords

# Denormalize output
denorm = np.array([w,w,h,h])
coords = (coords * denorm).astype(np.int32)
coords

# Daw bounding on top the image
xmin, xmax,ymin,ymax = coords[0]
pt1 =(xmin,ymin)
pt2 =(xmax,ymax)
print(pt1, pt2)
cv2.rectangle(image,pt1,pt2,(0,255,0),3)

plt.figure(figsize=(10,8))
plt.imshow(image)
plt.show()

Tensorflow Predictions

Detection Pipeline​

I can now wrap everything into a single function:

# Wrap all steps into a function
def object_detection(path):
    # Read image
    image = load_img(path) # PIL object
    image = np.array(image,dtype=np.uint8) # 8 bit array (0,255)
    image1 = load_img(path,target_size=(224,224))
    # Data preprocessing
    image_arr_224 = img_to_array(image1)/255.0  # convert into array and get the normalized output
    h,w,d = image.shape
    test_arr = image_arr_224.reshape(1,224,224,3)
    # Make predictions
    coords = model.predict(test_arr)
    # Denormalize the values
    denorm = np.array([w,w,h,h])
    coords = coords * denorm
    coords = coords.astype(np.int32)
    # Draw bounding on top the image
    xmin, xmax,ymin,ymax = coords[0]
    pt1 =(xmin,ymin)
    pt2 =(xmax,ymax)
    print(pt1, pt2)
    cv2.rectangle(image,pt1,pt2,(0,255,0),3)
    return image, coords

I can test the function on another image:

# Select another test image
path = './test_images/index3.jpg'
# Use function to select number plate
image, cods = object_detection(path)

plt.figure(figsize=(10,8))
plt.imshow(image)
plt.show()

Tensorflow Predictions