Vahe's picture
the digit extractor model changed
f5fea1b
import tensorflow as tf
import numpy as np
from PIL import Image
import cv2
def get_odometer_xy(model_path, image_path):
#model_path = 'odo_detector.tflite'
interpreter = tf.lite.Interpreter(model_path=model_path)
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
# Obtain the height and width of the corresponding image from the input tensor
image_height = input_details[0]['shape'][1] # 640
image_width = input_details[0]['shape'][2] # 640
# Image Preparation
# image_name = 'car.jpg'
image = Image.open(image_path)
image_resized = image.resize((image_width, image_height)) # Resize the image to the corresponding size of the input tensor and store it in a new variable
image_np = np.array(image_resized) #
image_np = np.true_divide(image_np, 255, dtype=np.float32)
image_np = image_np[np.newaxis, :]
# inference
interpreter.set_tensor(input_details[0]['index'], image_np)
interpreter.invoke()
# Obtaining output results
output = interpreter.get_tensor(output_details[0]['index'])
output = output[0]
output = output.T
boxes_xywh = output[:, :4] #Get coordinates of bounding box, first 4 columns of output tensor
scores = output[:, 4]#np.max(output[..., 5:], axis=1) #Get score value, 5th column of output tensor
classes = np.zeros(len(scores))#np.argmax(output[..., 5:], axis=1) # Get the class value, get the 6th and subsequent columns of the output tensor, and store the largest value in the output tensor.
# Threshold Setting
# threshold = 0.7
final_score = 0
x_center, y_center, width, height = 0, 0, 0, 0
class_name = 'odometer'
# Bounding boxes, scores, and classes are drawn on the image
# draw = ImageDraw.Draw(image_resized)
for box, score, cls in zip(boxes_xywh, scores, classes):
if score >= final_score:
x_center, y_center, width, height = box
final_score = score
class_name = cls
else:
pass
x1 = int((x_center - width / 2) * image_width)
y1 = int((y_center - height / 2) * image_height)
x2 = int((x_center + width / 2) * image_width)
y2 = int((y_center + height / 2) * image_height)
# draw.rectangle([x1, y1, x2, y2], outline="red", width=2)
# text = f"Class: {class_name}, Score: {final_score:.2f}"
# draw.text((x1, y1), text, fill="red")
# Saving Images
# image_resized.save('test_img.jpg')
return x1, y1, x2, y2, final_score
def get_digit(model_path, image_path, threshold=0.5):
interpreter = tf.lite.Interpreter(model_path=model_path)
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
# Obtain the height and width of the corresponding image from the input tensor
image_height = input_details[0]['shape'][1] # 640
image_width = input_details[0]['shape'][2] # 640
# Image Preparation
# image_name = 'car.jpg'
# image = Image.open(image_path2)
# image_resized = image.resize((image_width, image_height)) # Resize the image to the corresponding size of the input tensor and store it in a new variable
image = cv2.imread(image_path)
# image_resized = np.resize(image, (image_width, image_height, 3))
image_np = np.array(image) #
image_np = np.true_divide(image_np, 255, dtype=np.float32)
image_np = image_np[np.newaxis, :]
# inference
interpreter.set_tensor(input_details[0]['index'], image_np)
interpreter.invoke()
# Obtaining output results
output = interpreter.get_tensor(output_details[0]['index'])
output = output[0]
output = output.T
boxes_xywh = output[:, :4] #Get coordinates of bounding box, first 4 columns of output tensor
scores = np.max(output[:, 4:], axis=1) #Get score value, 5th column of output tensor
classes = np.argmax(output[:, 4:], axis=1) # Get the class value, get the 6th and subsequent columns of the output tensor, and store the largest value in the output tensor.
pred_list = []
prob_threshold = threshold
for box, score, cls in zip(boxes_xywh, scores, classes):
if score < prob_threshold:
continue
x_center, y_center, width, height = box
x1 = int((x_center - width / 2) * image_width)
y1 = int((y_center - height / 2) * image_height)
x2 = int((x_center + width / 2) * image_width)
y2 = int((y_center + height / 2) * image_height)
pred_list.append((x1, x2, cls, score))
pred_list = sorted(pred_list, key=lambda x: x[0])
num_list = []
temp_pred_list =[]
x_prev = 0
x_diff = min([elem[1] - elem[0] for elem in pred_list]) - 10
for idx, pred in enumerate(pred_list):
if idx == 0:
temp_pred_list.append(pred)
x_prev = pred[0]
elif idx == len(pred_list) - 1:
temp_final_num = sorted(temp_pred_list, key=lambda x: x[-1], reverse=True)[0]
num_list.append(temp_final_num)
elif pred[0] - x_prev < x_diff:
temp_pred_list.append(pred)
x_prev = pred[0]
else:
temp_final_num = sorted(temp_pred_list, key=lambda x: x[-1], reverse=True)[0]
num_list.append(temp_final_num)
temp_pred_list = []
x_prev = pred[0]
temp_pred_list.append(pred)
sorted_number_list = sorted(num_list, key=lambda x: x[0])
# sorted_number_list = sorted(sorted_number_list, reverse=True, key= lambda x: x[-1])
# output_digit = float(''.join([str(int(i[2])) if i[2]!=10 else '.' for i in sorted_number_list]))
output_digit = float(''.join([str(int(i[2])) if i[2]!=10 else '.' for i in sorted_number_list]))
# output_digit = ''.join([str(int(i[2])) if i[2]!=10 else '.' for i in sorted_number_list[:10]])
return output_digit