def test_step(batch_images, batch_labels, labels_list):
pred = crnn_model(batch_images, training=False)
loss_value = loss_object(y_true=batch_labels, y_pred=pred)
acc = test_accuracy(decoded_text=predict_text(pred),
true_label=labels_list)
loss_metric.update_state(values=loss_value)
test_accuracy_metric.update_state(values=acc)
def valid_step(batch_images, batch_labels):
pred = crnn_model(batch_images, training=False)
loss_value = loss(y_true=batch_labels, y_pred=pred)
valid_acc = accuracy(decoded_text=predict_text(
pred, blank_index=blank_index),
true_label=batch_labels)
valid_loss_metric.update_state(values=loss_value)
valid_accuracy.update_state(values=valid_acc)
def train_step(batch_images, batch_labels):
with tf.GradientTape() as tape:
pred = crnn_model(batch_images, training=True)
loss_value = loss(y_true=batch_labels, y_pred=pred)
gradients = tape.gradient(target=loss_value,
sources=crnn_model.trainable_variables)
optimizer.apply_gradients(
grads_and_vars=zip(gradients, crnn_model.trainable_variables))
train_loss_metric.update_state(values=loss_value)
train_acc = accuracy(decoded_text=predict_text(pred, blank_index),
true_label=batch_labels)
train_accuracy.update_state(values=train_acc)
def get_final_output_string(output, blank_index):
decoded_text = tf.cast(x=output, dtype=tf.int32)
decoded_text = decoded_text.numpy()
decoded_text = index_to_char(inputs=decoded_text, idx2char_dict=Config.get_idx2char(), blank_index=blank_index)
return decoded_text[0]
if __name__ == '__main__':
# GPU settings
gpus = tf.config.list_physical_devices("GPU")
if gpus:
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
num_classes, blank_index = get_num_classes_and_blank_index()
# read image
image_raw = tf.io.read_file(Config.test_picture_path)
image_tensor = tf.io.decode_image(contents=image_raw, channels=Config.IAMGE_CHANNELS, dtype=tf.dtypes.float32)
image_tensor = tf.image.resize(image_tensor, [Config.IMAGE_HEIGHT, Config.IMAGE_WIDTH])
image_tensor = tf.expand_dims(input=image_tensor, axis=0)
# load model
crnn_model = CRNN(num_classes)
crnn_model.load_weights(filepath=Config.save_model_dir+"saved_model")
pred = crnn_model(image_tensor, training=False)
predicted_string = get_final_output_string(predict_text(pred, blank_index), blank_index)
print(predicted_string)
output_tensor = tf.squeeze(output, axis=0)
idx2char = get_idx2char_map()
output_string_list = []
for i in range(output_tensor.shape[0]):
output_string_list.append(idx2char[output_tensor[i].numpy()])
output_string = "".join(output_string_list)
return output_string
if __name__ == '__main__':
# GPU settings
gpus = tf.config.list_physical_devices("GPU")
if gpus:
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
# read image
image_raw = tf.io.read_file(test_picture_path)
image_tensor = tf.io.decode_image(contents=image_raw,
channels=IAMGE_CHANNELS,
dtype=tf.dtypes.float32)
image_tensor = tf.image.resize(image_tensor, [IMAGE_HEIGHT, IMAGE_WIDTH])
image_tensor = tf.expand_dims(input=image_tensor, axis=0)
# load model
crnn_model = CRNN()
crnn_model.load_weights(filepath=save_model_dir + "saved_model")
pred = crnn_model(image_tensor, training=False)
predicted_string = get_final_output_string(predict_text(pred))
print(predicted_string)