tensorflow/efficientnet_b0

1import tensorflow as tf2import tensorflow_hub as hub3 4import requests5from PIL import Image6from io import BytesIO7 8import matplotlib.pyplot as plt9import numpy as np10import gradio as gr11 12#@title Helper functions for loading image (hidden)13 14original_image_cache = {}15 16def preprocess_image(image):17  image = np.array(image)18  # reshape into shape [batch_size, height, width, num_channels]19  img_reshaped = tf.reshape(image, [1, image.shape[0], image.shape[1], image.shape[2]])20  # Use `convert_image_dtype` to convert to floats in the [0,1] range.21  image = tf.image.convert_image_dtype(img_reshaped, tf.float32)22  return image23 24def load_image_from_url(img_url):25  """Returns an image with shape [1, height, width, num_channels]."""26  user_agent = {'User-agent': 'Colab Sample (https://tensorflow.org)'}27  response = requests.get(img_url, headers=user_agent)28  image = Image.open(BytesIO(response.content))29  image = preprocess_image(image)30  return image31 32def load_image(image_url, image_size=256, dynamic_size=False, max_dynamic_size=512):33  """Loads and preprocesses images."""34  # Cache image file locally.35  if image_url in original_image_cache:36    img = original_image_cache[image_url]37  elif image_url.startswith('https://'):38    img = load_image_from_url(image_url)39  else:40    fd = tf.io.gfile.GFile(image_url, 'rb')41    img = preprocess_image(Image.open(fd))42  original_image_cache[image_url] = img43  # Load and convert to float32 numpy array, add batch dimension, and normalize to range [0, 1].44  img_raw = img45  if tf.reduce_max(img) > 1.0:46    img = img / 255.47  if len(img.shape) == 3:48    img = tf.stack([img, img, img], axis=-1)49  if not dynamic_size:50    img = tf.image.resize_with_pad(img, image_size, image_size)51  elif img.shape[1] > max_dynamic_size or img.shape[2] > max_dynamic_size:52    img = tf.image.resize_with_pad(img, max_dynamic_size, max_dynamic_size)53  return img, img_raw54 55 56 57image_size = 22458dynamic_size = False59 60model_name = "efficientnet_b0" 61 62model_handle_map = {63  "efficientnetv2-s": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_s/classification/2",64  "efficientnetv2-m": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_m/classification/2",65  "efficientnetv2-l": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_l/classification/2",66  "efficientnetv2-s-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_s/classification/2",67  "efficientnetv2-m-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_m/classification/2",68  "efficientnetv2-l-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_l/classification/2",69  "efficientnetv2-xl-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_xl/classification/2",70  "efficientnetv2-b0-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b0/classification/2",71  "efficientnetv2-b1-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b1/classification/2",72  "efficientnetv2-b2-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b2/classification/2",73  "efficientnetv2-b3-21k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_b3/classification/2",74  "efficientnetv2-s-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_s/classification/2",75  "efficientnetv2-m-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_m/classification/2",76  "efficientnetv2-l-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_l/classification/2",77  "efficientnetv2-xl-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_xl/classification/2",78  "efficientnetv2-b0-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_b0/classification/2",79  "efficientnetv2-b1-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_b1/classification/2",80  "efficientnetv2-b2-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_b2/classification/2",81  "efficientnetv2-b3-21k-ft1k": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet21k_ft1k_b3/classification/2",82  "efficientnetv2-b0": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_b0/classification/2",83  "efficientnetv2-b1": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_b1/classification/2",84  "efficientnetv2-b2": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_b2/classification/2",85  "efficientnetv2-b3": "https://tfhub.dev/google/imagenet/efficientnet_v2_imagenet1k_b3/classification/2",86  "efficientnet_b0": "https://tfhub.dev/tensorflow/efficientnet/b0/classification/1",87  "efficientnet_b1": "https://tfhub.dev/tensorflow/efficientnet/b1/classification/1",88  "efficientnet_b2": "https://tfhub.dev/tensorflow/efficientnet/b2/classification/1",89  "efficientnet_b3": "https://tfhub.dev/tensorflow/efficientnet/b3/classification/1",90  "efficientnet_b4": "https://tfhub.dev/tensorflow/efficientnet/b4/classification/1",91  "efficientnet_b5": "https://tfhub.dev/tensorflow/efficientnet/b5/classification/1",92  "efficientnet_b6": "https://tfhub.dev/tensorflow/efficientnet/b6/classification/1",93  "efficientnet_b7": "https://tfhub.dev/tensorflow/efficientnet/b7/classification/1",94  "bit_s-r50x1": "https://tfhub.dev/google/bit/s-r50x1/ilsvrc2012_classification/1",95  "inception_v3": "https://tfhub.dev/google/imagenet/inception_v3/classification/4",96  "inception_resnet_v2": "https://tfhub.dev/google/imagenet/inception_resnet_v2/classification/4",97  "resnet_v1_50": "https://tfhub.dev/google/imagenet/resnet_v1_50/classification/4",98  "resnet_v1_101": "https://tfhub.dev/google/imagenet/resnet_v1_101/classification/4",99  "resnet_v1_152": "https://tfhub.dev/google/imagenet/resnet_v1_152/classification/4",100  "resnet_v2_50": "https://tfhub.dev/google/imagenet/resnet_v2_50/classification/4",101  "resnet_v2_101": "https://tfhub.dev/google/imagenet/resnet_v2_101/classification/4",102  "resnet_v2_152": "https://tfhub.dev/google/imagenet/resnet_v2_152/classification/4",103  "nasnet_large": "https://tfhub.dev/google/imagenet/nasnet_large/classification/4",104  "nasnet_mobile": "https://tfhub.dev/google/imagenet/nasnet_mobile/classification/4",105  "pnasnet_large": "https://tfhub.dev/google/imagenet/pnasnet_large/classification/4",106  "mobilenet_v2_100_224": "https://tfhub.dev/google/imagenet/mobilenet_v2_100_224/classification/4",107  "mobilenet_v2_130_224": "https://tfhub.dev/google/imagenet/mobilenet_v2_130_224/classification/4",108  "mobilenet_v2_140_224": "https://tfhub.dev/google/imagenet/mobilenet_v2_140_224/classification/4",109  "mobilenet_v3_small_100_224": "https://tfhub.dev/google/imagenet/mobilenet_v3_small_100_224/classification/5",110  "mobilenet_v3_small_075_224": "https://tfhub.dev/google/imagenet/mobilenet_v3_small_075_224/classification/5",111  "mobilenet_v3_large_100_224": "https://tfhub.dev/google/imagenet/mobilenet_v3_large_100_224/classification/5",112  "mobilenet_v3_large_075_224": "https://tfhub.dev/google/imagenet/mobilenet_v3_large_075_224/classification/5",113}114 115model_image_size_map = {116  "efficientnetv2-s": 384,117  "efficientnetv2-m": 480,118  "efficientnetv2-l": 480,119  "efficientnetv2-b0": 224,120  "efficientnetv2-b1": 240,121  "efficientnetv2-b2": 260,122  "efficientnetv2-b3": 300,123  "efficientnetv2-s-21k": 384,124  "efficientnetv2-m-21k": 480,125  "efficientnetv2-l-21k": 480,126  "efficientnetv2-xl-21k": 512,127  "efficientnetv2-b0-21k": 224,128  "efficientnetv2-b1-21k": 240,129  "efficientnetv2-b2-21k": 260,130  "efficientnetv2-b3-21k": 300,131  "efficientnetv2-s-21k-ft1k": 384,132  "efficientnetv2-m-21k-ft1k": 480,133  "efficientnetv2-l-21k-ft1k": 480,134  "efficientnetv2-xl-21k-ft1k": 512,135  "efficientnetv2-b0-21k-ft1k": 224,136  "efficientnetv2-b1-21k-ft1k": 240,137  "efficientnetv2-b2-21k-ft1k": 260,138  "efficientnetv2-b3-21k-ft1k": 300, 139  "efficientnet_b0": 224,140  "efficientnet_b1": 240,141  "efficientnet_b2": 260,142  "efficientnet_b3": 300,143  "efficientnet_b4": 380,144  "efficientnet_b5": 456,145  "efficientnet_b6": 528,146  "efficientnet_b7": 600,147  "inception_v3": 299,148  "inception_resnet_v2": 299,149  "mobilenet_v2_100_224": 224,150  "mobilenet_v2_130_224": 224,151  "mobilenet_v2_140_224": 224,152  "nasnet_large": 331,153  "nasnet_mobile": 224,154  "pnasnet_large": 331,155  "resnet_v1_50": 224,156  "resnet_v1_101": 224,157  "resnet_v1_152": 224,158  "resnet_v2_50": 224,159  "resnet_v2_101": 224,160  "resnet_v2_152": 224,161  "mobilenet_v3_small_100_224": 224,162  "mobilenet_v3_small_075_224": 224,163  "mobilenet_v3_large_100_224": 224,164  "mobilenet_v3_large_075_224": 224,165}166 167model_handle = model_handle_map[model_name]168 169 170max_dynamic_size = 512171if model_name in model_image_size_map:172  image_size = model_image_size_map[model_name]173  dynamic_size = False174  print(f"Images will be converted to {image_size}x{image_size}")175else:176  dynamic_size = True177  print(f"Images will be capped to a max size of {max_dynamic_size}x{max_dynamic_size}")178 179labels_file = "https://storage.googleapis.com/download.tensorflow.org/data/ImageNetLabels.txt"180 181#download labels and creates a maps182downloaded_file = tf.keras.utils.get_file("labels.txt", origin=labels_file)183 184classes = []185 186with open(downloaded_file) as f:187  labels = f.readlines()188  classes = [l.strip() for l in labels]189 190 191classifier = hub.load(model_handle)192 193 194def inference(img):195  image, original_image = load_image(img, image_size, dynamic_size, max_dynamic_size)  196  197  198  input_shape = image.shape199  warmup_input = tf.random.uniform(input_shape, 0, 1.0)200  warmup_logits = classifier(warmup_input).numpy()201  202  # Run model on image203  probabilities = tf.nn.softmax(classifier(image)).numpy()204  205  top_5 = tf.argsort(probabilities, axis=-1, direction="DESCENDING")[0][:5].numpy()206  np_classes = np.array(classes)207  208  # Some models include an additional 'background' class in the predictions, so209  # we must account for this when reading the class labels.210  includes_background_class = probabilities.shape[1] == 1001211  result = {}212  for i, item in enumerate(top_5):213    class_index = item if includes_background_class else item + 1214    line = f'({i+1}) {class_index:4} - {classes[class_index]}: {probabilities[0][top_5][i]}'215    result[classes[class_index]] = probabilities[0][top_5][i].item()216  return result217 218title="efficientnet_b0"219description="Gradio Demo for efficientnet_b0: Imagenet (ILSVRC-2012-CLS) classification with EfficientNet-B0. To use it, simply upload your image or click on one of the examples to load them. Read more at the links below"220 221article = "<p style='text-align: center'><a href='https://tfhub.dev/google/efficientnet/b0/classification/1' target='_blank'>Tensorflow Hub</a></p>"222examples=[['apple1.jpg']]223 224gr.Interface(inference,gr.inputs.Image(type="filepath"),"label",title=title,description=description,article=article,examples=examples).launch(enable_queue=True)