Код: Выделить всё
imgs_dir = r'C:\Users\danie\sign-language-alpha\data\asl_alphabet_test\asl_alphabet_test'
imgs = os.listdir(imgs_dir)
class_mapping = train_images.class_indices
def get_class_label(predictions, class_mapping):
labels_mapping = {v: k for k, v in class_mapping.items()}
predicted_labels = [labels_mapping[pred] for pred in predictions]
return predicted_labels
def predict_image_class(model, img_path, class_mapping):
img = cv2.imread(os.path.join(imgs_dir, img_path))
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img = np.expand_dims(cv2.resize(img, (64, 64)), axis = 0)
predictions = np.argmax(model.predict(img, verbose = 0), axis = 1)
predicted_labels = get_class_label(predictions, class_mapping)
return predicted_labels
import cv2
import numpy as np
def process_and_save_video_frame(frame, save_path="processed_frame.jpg"):
"""
Process a video frame: convert to grayscale, resize with aspect ratio,
normalize, reshape, and save the processed frame.
Args:
frame: The original video frame captured from the webcam.
save_path: The path to save the processed frame image.
Returns:
frame_final: The processed frame ready for model input.
"""
def resize_with_aspect_ratio(image, target_size):
h, w = image.shape[:2]
target_w, target_h = target_size
scale = min(target_w / w, target_h / h)
new_w, new_h = int(w * scale), int(h * scale)
resized = cv2.resize(image, (new_w, new_h))
delta_w, delta_h = target_w - new_w, target_h - new_h
top, bottom = delta_h // 2, delta_h - (delta_h // 2)
left, right = delta_w // 2, delta_w - (delta_w // 2)
color = [0, 0, 0]
new_image = cv2.copyMakeBorder(resized, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)
return new_image
print(f"Original Frame shape: {frame.shape}")
# Convert to grayscale
frame_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# Resize with aspect ratio
frame_resized = resize_with_aspect_ratio(frame_gray, (64, 64))
print(f"Shape after resizing with aspect ratio to 64x64: {frame_resized.shape}")
# Normalize
frame_normalized = frame_resized.astype('float32') / 255.0
# Add batch and channel dimensions
frame_final = np.expand_dims(frame_normalized, axis=0)
frame_final = np.expand_dims(frame_final, axis=-1)
print(f"Shape after adding batch and channel dimensions: {frame_final.shape}")
# Save the processed frame
cv2.imwrite(save_path, frame_resized * 255)
return frame_final
for img in imgs:
predicted_labels = predict_image_class(model, img, class_mapping)
print(f'Predicted Labels {img} -----> {predicted_labels}')
Код: Выделить всё
import os
import cv2
import numpy as np
from keras.models import load_model
model = load_model("FINAL.h5")
imgs_dir = r'C:\Users\danie\sign-language-alpha\data\asl_alphabet_test\asl_alphabet_test'
imgs = os.listdir(imgs_dir)
class_mapping = train_images.class_indices
print(f"Class mapping: {class_mapping}")
def get_class_label(predictions, class_mapping):
labels_mapping = {v: k for k, v in class_mapping.items()}
predicted_labels = [labels_mapping[pred] for pred in predictions]
print(predicted_labels)
return predicted_labels
def predict_image_class_from_path(model, img_path, class_mapping):
img = cv2.imread(img_path)
if img is None:
print(f"Error: Unable to load image at path {img_path}")
return ["Error"]
print(f"Original image shape: {img.shape}")
# Convert to grayscale
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
print(f"Shape after converting to grayscale: {img.shape}")
# Resize the image to 64x64
img = cv2.resize(img, (64, 64))
print(f"Shape after resizing to 64x64: {img.shape}")
# Normalize the image
img = img.astype('float32') / 255.0
# Add batch dimension and channel dimension
img = np.expand_dims(img, axis=0)
img = np.expand_dims(img, axis=-1)
print(f"Shape after adding batch and channel dimensions: {img.shape}")
# Make predictions
predictions = np.argmax(model.predict(img, verbose=0), axis=1)
print(f"Predictions: {predictions}")
# Get the predicted label
predicted_labels = get_class_label(predictions, class_mapping)
return predicted_labels
def predict_image_class_from_frame(model, frame, class_mapping):
# Convert to grayscale
img = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
print(f"Shape after converting to grayscale: {img.shape}")
# Resize the image to 64x64
img = cv2.resize(img, (64, 64))
print(f"Shape after resizing to 64x64: {img.shape}")
# Normalize the image
img = img.astype('float32') / 255.0
# Add batch dimension and channel dimension
img = np.expand_dims(img, axis=0)
img = np.expand_dims(img, axis=-1)
print(f"Shape after adding batch and channel dimensions: {img.shape}")
# Make predictions
predictions = np.argmax(model.predict(img, verbose=0), axis=1)
print(f"Predictions: {predictions}")
# Get the predicted label
predicted_labels = get_class_label(predictions, class_mapping)
return predicted_labels
cap = cv2.VideoCapture(0)
while True:
ret, frame = cap.read()
if not ret:
break
frame_resized = cv2.resize(frame, (200, 200))
predicted_labels = predict_image_class_from_frame(model, frame_resized, class_mapping)
# Display the predicted label on the frame
cv2.putText(frame_resized, predicted_labels[0], (50, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2, cv2.LINE_AA)
# Display the frame
cv2.imshow('Sign Language Prediction', frame_resized)
# Break the loop on 'q' key press
if cv2.waitKey(1) & 0xFF == ord('q'):
break
#{'A': 0, 'B': 1, 'C': 2, 'D': 3, 'E': 4, 'F': 5, 'G': 6, 'H': 7, 'I': 8, 'J': 9, 'K': 10, 'L': 11, 'M': 12, 'N': 13, 'O': 14, 'P': 15, 'Q': 16, 'R': 17, 'S': 18, 'T': 19, 'U': 20, 'V': 21, 'W': 22, 'X': 23, 'Y': 24, 'Z': 25, 'del': 26, 'nothing': 27, 'space': 28}
cap.release()
cv2.destroyAllWindows()
Подробнее здесь: https://stackoverflow.com/questions/787 ... ning-model
