Tutorial: Executar um modelo TensorFlow em Python

Este tutorial mostra como usar um modelo TensorFlow exportado localmente para classificar imagens.

Nota

Este tutorial aplica-se apenas a modelos exportados de projetos de classificação de imagem "Geral (compacto)". Se você exportou outros modelos, visite nosso repositório de código de exemplo.

Pré-requisitos

  • Instale o Python 2.7+ ou Python 3.6+.
  • Instale o pip.

Em seguida, você precisará instalar os seguintes pacotes:

pip install tensorflow
pip install pillow
pip install numpy
pip install opencv-python

Carregar o modelo e as etiquetas

O arquivo de .zip baixado da etapa de exportação contém um arquivo model.pb e um arquivo labels.txt. Estes ficheiros representam o modelo preparado e as etiquetas de classificação. O primeiro passo é carregar o modelo para o seu projeto. Adicione o seguinte código a um novo script Python.

import tensorflow as tf
import os

graph_def = tf.compat.v1.GraphDef()
labels = []

# These are set to the default names from exported models, update as needed.
filename = "model.pb"
labels_filename = "labels.txt"

# Import the TF graph
with tf.io.gfile.GFile(filename, 'rb') as f:
    graph_def.ParseFromString(f.read())
    tf.import_graph_def(graph_def, name='')

# Create a list of labels.
with open(labels_filename, 'rt') as lf:
    for l in lf:
        labels.append(l.strip())

Preparar uma imagem para predição

Há algumas etapas que você precisa seguir para preparar uma imagem para previsão. Estes passos imitam a manipulação de imagem realizada durante o treino.

  1. Abrir o ficheiro e criar uma imagem na colorimetria BGR

    from PIL import Image
    import numpy as np
    import cv2
    
    # Load from a file
    imageFile = "<path to your image file>"
    image = Image.open(imageFile)
    
    # Update orientation based on EXIF tags, if the file has orientation info.
    image = update_orientation(image)
    
    # Convert to OpenCV format
    image = convert_to_opencv(image)
    
  2. Se a imagem tiver uma dimensão maior que 1600 pixels, chame esse método (definido posteriormente).

    image = resize_down_to_1600_max_dim(image)
    
  3. Recortar o quadrado central maior

    h, w = image.shape[:2]
    min_dim = min(w,h)
    max_square_image = crop_center(image, min_dim, min_dim)
    
  4. Redimensione esse quadrado para 256x256

    augmented_image = resize_to_256_square(max_square_image)
    
  5. Recortar o centro para o tamanho de entrada específico para o modelo

    # Get the input size of the model
    with tf.compat.v1.Session() as sess:
        input_tensor_shape = sess.graph.get_tensor_by_name('Placeholder:0').shape.as_list()
    network_input_size = input_tensor_shape[1]
    
    # Crop the center for the specified network_input_Size
    augmented_image = crop_center(augmented_image, network_input_size, network_input_size)
    
    
  6. Defina funções auxiliares. Os passos acima utilizam as seguintes funções de programa auxiliar:

    def convert_to_opencv(image):
        # RGB -> BGR conversion is performed as well.
        image = image.convert('RGB')
        r,g,b = np.array(image).T
        opencv_image = np.array([b,g,r]).transpose()
        return opencv_image
    
    def crop_center(img,cropx,cropy):
        h, w = img.shape[:2]
        startx = w//2-(cropx//2)
        starty = h//2-(cropy//2)
        return img[starty:starty+cropy, startx:startx+cropx]
    
    def resize_down_to_1600_max_dim(image):
        h, w = image.shape[:2]
        if (h < 1600 and w < 1600):
            return image
    
        new_size = (1600 * w // h, 1600) if (h > w) else (1600, 1600 * h // w)
        return cv2.resize(image, new_size, interpolation = cv2.INTER_LINEAR)
    
    def resize_to_256_square(image):
        h, w = image.shape[:2]
        return cv2.resize(image, (256, 256), interpolation = cv2.INTER_LINEAR)
    
    def update_orientation(image):
        exif_orientation_tag = 0x0112
        if hasattr(image, '_getexif'):
            exif = image._getexif()
            if (exif != None and exif_orientation_tag in exif):
                orientation = exif.get(exif_orientation_tag, 1)
                # orientation is 1 based, shift to zero based and flip/transpose based on 0-based values
                orientation -= 1
                if orientation >= 4:
                    image = image.transpose(Image.TRANSPOSE)
                if orientation == 2 or orientation == 3 or orientation == 6 or orientation == 7:
                    image = image.transpose(Image.FLIP_TOP_BOTTOM)
                if orientation == 1 or orientation == 2 or orientation == 5 or orientation == 6:
                    image = image.transpose(Image.FLIP_LEFT_RIGHT)
        return image
    

Classificar uma imagem

Uma vez que a imagem é preparada como um tensor, podemos enviá-la através do modelo para uma previsão.

# These names are part of the model and cannot be changed.
output_layer = 'loss:0'
input_node = 'Placeholder:0'

with tf.compat.v1.Session() as sess:
    try:
        prob_tensor = sess.graph.get_tensor_by_name(output_layer)
        predictions = sess.run(prob_tensor, {input_node: [augmented_image] })
    except KeyError:
        print ("Couldn't find classification output layer: " + output_layer + ".")
        print ("Verify this a model exported from an Object Detection project.")
        exit(-1)

Apresentar os resultados

Os resultados da execução do tensor de imagens através do modelo terão de ser mapeados novamente para as etiquetas.

    # Print the highest probability label
    highest_probability_index = np.argmax(predictions)
    print('Classified as: ' + labels[highest_probability_index])
    print()

    # Or you can print out all of the results mapping labels to probabilities.
    label_index = 0
    for p in predictions:
        truncated_probablity = np.float64(np.round(p,8))
        print (labels[label_index], truncated_probablity)
        label_index += 1

Próximos passos

Em seguida, saiba como envolver seu modelo em um aplicativo móvel: