Path: blob/master/site/pt-br/hub/tutorials/image_classification.ipynb
38471 views
Kernel: Python 3
Copyright 2021 The TensorFlow Hub Authors.
Licensed under the Apache License, Version 2.0 (the "License");
In [ ]:
# Copyright 2021 The TensorFlow Hub Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ==============================================================================
Classificação de imagens com o TensorFlow Hub
Neste Colab, você testará diversos modelos de classificação do TensorFlow Hub e decidirá qual é o melhor para seu caso de uso.
Como o TF Hub incentiva uma convenção consistente de entradas para modelos que fazem operações com imagens, é fácil testar diferentes arquiteturas para encontrar a mais adequada às suas necessidades.
Ver em TensorFlow.org
Executar no Google Colab
Ver no GitHub
Baixar notebook
Ver modelos do TF HubIn [ ]:
import tensorflow as tf import tensorflow_hub as hub import requests from PIL import Image from io import BytesIO import matplotlib.pyplot as plt import numpy as np
In [ ]:
#@title Helper functions for loading image (hidden) original_image_cache = {} def preprocess_image(image): image = np.array(image) # reshape into shape [batch_size, height, width, num_channels] img_reshaped = tf.reshape(image, [1, image.shape[0], image.shape[1], image.shape[2]]) # Use `convert_image_dtype` to convert to floats in the [0,1] range. image = tf.image.convert_image_dtype(img_reshaped, tf.float32) return image def load_image_from_url(img_url): """Returns an image with shape [1, height, width, num_channels].""" user_agent = {'User-agent': 'Colab Sample (https://tensorflow.org)'} response = requests.get(img_url, headers=user_agent) image = Image.open(BytesIO(response.content)) image = preprocess_image(image) return image def load_image(image_url, image_size=256, dynamic_size=False, max_dynamic_size=512): """Loads and preprocesses images.""" # Cache image file locally. if image_url in original_image_cache: img = original_image_cache[image_url] elif image_url.startswith('https://'): img = load_image_from_url(image_url) else: fd = tf.io.gfile.GFile(image_url, 'rb') img = preprocess_image(Image.open(fd)) original_image_cache[image_url] = img # Load and convert to float32 numpy array, add batch dimension, and normalize to range [0, 1]. img_raw = img if tf.reduce_max(img) > 1.0: img = img / 255. if len(img.shape) == 3: img = tf.stack([img, img, img], axis=-1) if not dynamic_size: img = tf.image.resize_with_pad(img, image_size, image_size) elif img.shape[1] > max_dynamic_size or img.shape[2] > max_dynamic_size: img = tf.image.resize_with_pad(img, max_dynamic_size, max_dynamic_size) return img, img_raw def show_image(image, title=''): image_size = image.shape[1] w = (image_size * 6) // 320 plt.figure(figsize=(w, w)) plt.imshow(image[0], aspect='equal') plt.axis('off') plt.title(title) plt.show()
Selecione um modelo de classificação de imagens. Em seguida, algumas variáveis são definidas, e o arquivo de rótulos é baixado e preparado para uso.
Existem algumas diferenças técnicas entre os modelos, como tamanho de entrada, tamanho do modelo, exatidão e tempo de inferência diferentes. Aqui, você pode alterar o modelo usado até encontrar o mais adequado para seu caso de uso.
O identificador (URL) do modelo é exibido via print por questões de conveniência. Confira mais documentações sobre cada modelo nessas URLs.
Observação: todos esses modelos foram treinados com o dataset ImageNet.
In [ ]:
#@title Select an Image Classification model image_size = 224 dynamic_size = False model_name = "efficientnetv2-s" # @param ['efficientnetv2-s', 'efficientnetv2-m', 'efficientnetv2-l', 'efficientnetv2-s-21k', 'efficientnetv2-m-21k', 'efficientnetv2-l-21k', 'efficientnetv2-xl-21k', 'efficientnetv2-b0-21k', 'efficientnetv2-b1-21k', 'efficientnetv2-b2-21k', 'efficientnetv2-b3-21k', 'efficientnetv2-s-21k-ft1k', 'efficientnetv2-m-21k-ft1k', 'efficientnetv2-l-21k-ft1k', 'efficientnetv2-xl-21k-ft1k', 'efficientnetv2-b0-21k-ft1k', 'efficientnetv2-b1-21k-ft1k', 'efficientnetv2-b2-21k-ft1k', 'efficientnetv2-b3-21k-ft1k', 'efficientnetv2-b0', 'efficientnetv2-b1', 'efficientnetv2-b2', 'efficientnetv2-b3', 'efficientnet_b0', 'efficientnet_b1', 'efficientnet_b2', 'efficientnet_b3', 'efficientnet_b4', 'efficientnet_b5', 'efficientnet_b6', 'efficientnet_b7', 'bit_s-r50x1', 'inception_v3', 'inception_resnet_v2', 'resnet_v1_50', 'resnet_v1_101', 'resnet_v1_152', 'resnet_v2_50', 'resnet_v2_101', 'resnet_v2_152', 'nasnet_large', 'nasnet_mobile', 'pnasnet_large', 'mobilenet_v2_100_224', 'mobilenet_v2_130_224', 'mobilenet_v2_140_224', 'mobilenet_v3_small_100_224', 'mobilenet_v3_small_075_224', 'mobilenet_v3_large_100_224', 'mobilenet_v3_large_075_224'] model_handle_map = { "efficientnetv2-s": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_s/classification/2", "efficientnetv2-m": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_m/classification/2", "efficientnetv2-l": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_l/classification/2", "efficientnetv2-s-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_s/classification/2", "efficientnetv2-m-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_m/classification/2", "efficientnetv2-l-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_l/classification/2", "efficientnetv2-xl-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_xl/classification/2", "efficientnetv2-b0-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b0/classification/2", "efficientnetv2-b1-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b1/classification/2", "efficientnetv2-b2-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b2/classification/2", "efficientnetv2-b3-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b3/classification/2", "efficientnetv2-s-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_s/classification/2", "efficientnetv2-m-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_m/classification/2", "efficientnetv2-l-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_l/classification/2", "efficientnetv2-xl-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_xl/classification/2", "efficientnetv2-b0-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_b0/classification/2", "efficientnetv2-b1-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_b1/classification/2", "efficientnetv2-b2-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_b2/classification/2", "efficientnetv2-b3-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_b3/classification/2", "efficientnetv2-b0": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_b0/classification/2", "efficientnetv2-b1": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_b1/classification/2", "efficientnetv2-b2": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_b2/classification/2", "efficientnetv2-b3": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_b3/classification/2", "efficientnet_b0": "https://tfhub.dev/tensorflow/efficientnet/b0/classification/1", "efficientnet_b1": "https://tfhub.dev/tensorflow/efficientnet/b1/classification/1", "efficientnet_b2": "https://tfhub.dev/tensorflow/efficientnet/b2/classification/1", "efficientnet_b3": "https://tfhub.dev/tensorflow/efficientnet/b3/classification/1", "efficientnet_b4": "https://tfhub.dev/tensorflow/efficientnet/b4/classification/1", "efficientnet_b5": "https://tfhub.dev/tensorflow/efficientnet/b5/classification/1", "efficientnet_b6": "https://tfhub.dev/tensorflow/efficientnet/b6/classification/1", "efficientnet_b7": "https://tfhub.dev/tensorflow/efficientnet/b7/classification/1", "bit_s-r50x1": "https://tfhub.dev/google/bit/s-r50x1/ilsvrc2012_classification/1", "inception_v3": "https://tfhub.dev/google/imagenet/inception_v3/classification/4", "inception_resnet_v2": "https://tfhub.dev/google/imagenet/inception_resnet_v2/classification/4", "resnet_v1_50": "https://tfhub.dev/google/imagenet/resnet_v1_50/classification/4", "resnet_v1_101": "https://tfhub.dev/google/imagenet/resnet_v1_101/classification/4", "resnet_v1_152": "https://tfhub.dev/google/imagenet/resnet_v1_152/classification/4", "resnet_v2_50": "https://tfhub.dev/google/imagenet/resnet_v2_50/classification/4", "resnet_v2_101": "https://tfhub.dev/google/imagenet/resnet_v2_101/classification/4", "resnet_v2_152": "https://tfhub.dev/google/imagenet/resnet_v2_152/classification/4", "nasnet_large": "https://tfhub.dev/google/imagenet/nasnet_large/classification/4", "nasnet_mobile": "https://tfhub.dev/google/imagenet/nasnet_mobile/classification/4", "pnasnet_large": "https://tfhub.dev/google/imagenet/pnasnet_large/classification/4", "mobilenet_v2_100_224": "https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/classification/4", "mobilenet_v2_130_224": "https://tfhub.dev/google/imagenet/mobilenet_v2_130_224/classification/4", "mobilenet_v2_140_224": "https://tfhub.dev/google/imagenet/mobilenet_v2_140_224/classification/4", "mobilenet_v3_small_100_224": "https://tfhub.dev/google/imagenet/mobilenet_v3_small_100_224/classification/5", "mobilenet_v3_small_075_224": "https://tfhub.dev/google/imagenet/mobilenet_v3_small_075_224/classification/5", "mobilenet_v3_large_100_224": "https://tfhub.dev/google/imagenet/mobilenet_v3_large_100_224/classification/5", "mobilenet_v3_large_075_224": "https://tfhub.dev/google/imagenet/mobilenet_v3_large_075_224/classification/5", } model_image_size_map = { "efficientnetv2-s": 384, "efficientnetv2-m": 480, "efficientnetv2-l": 480, "efficientnetv2-b0": 224, "efficientnetv2-b1": 240, "efficientnetv2-b2": 260, "efficientnetv2-b3": 300, "efficientnetv2-s-21k": 384, "efficientnetv2-m-21k": 480, "efficientnetv2-l-21k": 480, "efficientnetv2-xl-21k": 512, "efficientnetv2-b0-21k": 224, "efficientnetv2-b1-21k": 240, "efficientnetv2-b2-21k": 260, "efficientnetv2-b3-21k": 300, "efficientnetv2-s-21k-ft1k": 384, "efficientnetv2-m-21k-ft1k": 480, "efficientnetv2-l-21k-ft1k": 480, "efficientnetv2-xl-21k-ft1k": 512, "efficientnetv2-b0-21k-ft1k": 224, "efficientnetv2-b1-21k-ft1k": 240, "efficientnetv2-b2-21k-ft1k": 260, "efficientnetv2-b3-21k-ft1k": 300, "efficientnet_b0": 224, "efficientnet_b1": 240, "efficientnet_b2": 260, "efficientnet_b3": 300, "efficientnet_b4": 380, "efficientnet_b5": 456, "efficientnet_b6": 528, "efficientnet_b7": 600, "inception_v3": 299, "inception_resnet_v2": 299, "mobilenet_v2_100_224": 224, "mobilenet_v2_130_224": 224, "mobilenet_v2_140_224": 224, "nasnet_large": 331, "nasnet_mobile": 224, "pnasnet_large": 331, "resnet_v1_50": 224, "resnet_v1_101": 224, "resnet_v1_152": 224, "resnet_v2_50": 224, "resnet_v2_101": 224, "resnet_v2_152": 224, "mobilenet_v3_small_100_224": 224, "mobilenet_v3_small_075_224": 224, "mobilenet_v3_large_100_224": 224, "mobilenet_v3_large_075_224": 224, } model_handle = model_handle_map[model_name] print(f"Selected model: {model_name} : {model_handle}") max_dynamic_size = 512 if model_name in model_image_size_map: image_size = model_image_size_map[model_name] dynamic_size = False print(f"Images will be converted to {image_size}x{image_size}") else: dynamic_size = True print(f"Images will be capped to a max size of {max_dynamic_size}x{max_dynamic_size}") labels_file = "https://storage.googleapis.com/download.tensorflow.org/data/ImageNetLabels.txt" #download labels and creates a maps downloaded_file = tf.keras.utils.get_file("labels.txt", origin=labels_file) classes = [] with open(downloaded_file) as f: labels = f.readlines() classes = [l.strip() for l in labels]
Você pode selecionar uma das imagens abaixo ou usar sua própria imagem, mas lembre-se de que o tamanho da entrada dos modelos varia, e alguns modelos usam um tamanho de entrada dinâmico (permitindo inferência com uma imagem não dimensionada). Dito isso, o método load_image já faz o redimensionamento da imagem para o formato esperado.
In [ ]:
#@title Select an Input Image image_name = "turtle" # @param ['tiger', 'bus', 'car', 'cat', 'dog', 'apple', 'banana', 'turtle', 'flamingo', 'piano', 'honeycomb', 'teapot'] images_for_test_map = { "tiger": "https://upload.wikimedia.org/wikipedia/commons/b/b0/Bengal_tiger_%28Panthera_tigris_tigris%29_female_3_crop.jpg", #by Charles James Sharp, CC BY-SA 4.0 <https://creativecommons.org/licenses/by-sa/4.0>, via Wikimedia Commons "bus": "https://upload.wikimedia.org/wikipedia/commons/6/63/LT_471_%28LTZ_1471%29_Arriva_London_New_Routemaster_%2819522859218%29.jpg", #by Martin49 from London, England, CC BY 2.0 <https://creativecommons.org/licenses/by/2.0>, via Wikimedia Commons "car": "https://upload.wikimedia.org/wikipedia/commons/4/49/2013-2016_Toyota_Corolla_%28ZRE172R%29_SX_sedan_%282018-09-17%29_01.jpg", #by EurovisionNim, CC BY-SA 4.0 <https://creativecommons.org/licenses/by-sa/4.0>, via Wikimedia Commons "cat": "https://upload.wikimedia.org/wikipedia/commons/4/4d/Cat_November_2010-1a.jpg", #by Alvesgaspar, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons "dog": "https://upload.wikimedia.org/wikipedia/commons/archive/a/a9/20090914031557%21Saluki_dog_breed.jpg", #by Craig Pemberton, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons "apple": "https://upload.wikimedia.org/wikipedia/commons/1/15/Red_Apple.jpg", #by Abhijit Tembhekar from Mumbai, India, CC BY 2.0 <https://creativecommons.org/licenses/by/2.0>, via Wikimedia Commons "banana": "https://upload.wikimedia.org/wikipedia/commons/1/1c/Bananas_white_background.jpg", #by fir0002 flagstaffotos [at] gmail.com Canon 20D + Tamron 28-75mm f/2.8, GFDL 1.2 <http://www.gnu.org/licenses/old-licenses/fdl-1.2.html>, via Wikimedia Commons "turtle": "https://upload.wikimedia.org/wikipedia/commons/8/80/Turtle_golfina_escobilla_oaxaca_mexico_claudio_giovenzana_2010.jpg", #by Claudio Giovenzana, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons "flamingo": "https://upload.wikimedia.org/wikipedia/commons/b/b8/James_Flamingos_MC.jpg", #by Christian Mehlführer, User:Chmehl, CC BY 3.0 <https://creativecommons.org/licenses/by/3.0>, via Wikimedia Commons "piano": "https://upload.wikimedia.org/wikipedia/commons/d/da/Steinway_%26_Sons_upright_piano%2C_model_K-132%2C_manufactured_at_Steinway%27s_factory_in_Hamburg%2C_Germany.png", #by "Photo: © Copyright Steinway & Sons", CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons "honeycomb": "https://upload.wikimedia.org/wikipedia/commons/f/f7/Honey_comb.jpg", #by Merdal, CC BY-SA 3.0 <http://creativecommons.org/licenses/by-sa/3.0/>, via Wikimedia Commons "teapot": "https://upload.wikimedia.org/wikipedia/commons/4/44/Black_tea_pot_cropped.jpg", #by Mendhak, CC BY-SA 2.0 <https://creativecommons.org/licenses/by-sa/2.0>, via Wikimedia Commons } img_url = images_for_test_map[image_name] image, original_image = load_image(img_url, image_size, dynamic_size, max_dynamic_size) show_image(image, 'Scaled image')
Agora que o modelo foi escolhido, é simples carregá-lo com o TensorFlow Hub.
O modelo também é chamado com uma entrada aleatória como uma execução "de preparação". Chamadas subsequentes costumam ser muito mais rápidas, e você pode comparar com a latência abaixo.
Observação: os modelos que usam um tamanho dinâmico podem precisar de uma nova execução "de preparação" para cada tamanho de imagem.
In [ ]:
classifier = hub.load(model_handle) input_shape = image.shape warmup_input = tf.random.uniform(input_shape, 0, 1.0) %time warmup_logits = classifier(warmup_input).numpy()
Está tudo pronto para a inferência. Aqui você pode ver os 5 primeiros resultados do modelo para a imagem selecionada.
In [ ]:
# Run model on image %time probabilities = tf.nn.softmax(classifier(image)).numpy() top_5 = tf.argsort(probabilities, axis=-1, direction="DESCENDING")[0][:5].numpy() np_classes = np.array(classes) # Some models include an additional 'background' class in the predictions, so # we must account for this when reading the class labels. includes_background_class = probabilities.shape[1] == 1001 for i, item in enumerate(top_5): class_index = item if includes_background_class else item + 1 line = f'({i+1}) {class_index:4} - {classes[class_index]}: {probabilities[0][top_5][i]}' print(line) show_image(image, '')
Saiba mais
Se você quiser saber mais e tentar fazer aprendizado por transferência com esses modelos, confira este tutorial: Aprendizado por transferência para classificação de imagens.
Se você quiser verificar mais modelos de imagens, confira em tfhub.dev.