Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
hackassin
GitHub Repository: hackassin/learnopencv
 Path: blob/master/EfficientNet/EfficientNet_ResNet50.ipynb
3118 views
Kernel: Python 3

#EfficientNet Inference in Keras

!pip install efficientnet
Collecting efficientnet Downloading https://files.pythonhosted.org/packages/66/9c/8bee2927037da1a56e899eb62140fba2862a0d4faeeacbdbe7a1f27bd528/efficientnet-0.0.3-py2.py3-none-any.whl Installing collected packages: efficientnet Successfully installed efficientnet-0.0.3 
from keras.applications.imagenet_utils import decode_predictions
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
Using TensorFlow backend.

EfficientNet Model Size

  1. EfficientNet-B0: 20.8 MB

  2. EfficientNet-B1: 30.7 MB

  3. EfficientNet-B2: 35.7 MB

  4. EfficientNet-B3: 47.8 MB

Let's go for the smallest model of the above 4 : EfficienNet-B0.

from efficientnet import EfficientNetB0

model = EfficientNetB0(weights='imagenet')
WARNING: Logging before flag parsing goes to stderr. W0701 03:55:41.854610 139951651768192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:74: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead. W0701 03:55:41.909897 139951651768192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:517: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead. W0701 03:55:41.919556 139951651768192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/efficientnet/initializers.py:30: The name tf.random_normal is deprecated. Please use tf.random.normal instead. W0701 03:55:41.975130 139951651768192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:174: The name tf.get_default_session is deprecated. Please use tf.compat.v1.get_default_session instead. W0701 03:55:41.980821 139951651768192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:181: The name tf.ConfigProto is deprecated. Please use tf.compat.v1.ConfigProto instead. W0701 03:55:42.291757 139951651768192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:1834: The name tf.nn.fused_batch_norm is deprecated. Please use tf.compat.v1.nn.fused_batch_norm instead. W0701 03:55:43.210394 139951651768192 deprecation.py:323] From /usr/local/lib/python3.6/dist-packages/efficientnet/layers.py:29: div (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version. Instructions for updating: Deprecated in favor of operator or tf.math.divide. W0701 03:55:47.900598 139951651768192 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3445: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version. Instructions for updating: Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.
Downloading data from https://github.com/qubvel/efficientnet/releases/download/v0.0.1/efficientnet-b0_imagenet_1000.h5 21856256/21849008 [==============================] - 0s 0us/step 
# Download the test image
!wget https://github.com/qubvel/efficientnet/raw/master/misc/panda.jpg
--2019-07-01 03:55:53-- https://github.com/qubvel/efficientnet/raw/master/misc/panda.jpg Resolving github.com (github.com)... 192.30.253.113 Connecting to github.com (github.com)|192.30.253.113|:443... connected. HTTP request sent, awaiting response... 302 Found Location: https://raw.githubusercontent.com/qubvel/efficientnet/master/misc/panda.jpg [following] --2019-07-01 03:55:53-- https://raw.githubusercontent.com/qubvel/efficientnet/master/misc/panda.jpg Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.0.133, 151.101.64.133, 151.101.128.133, ... Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.0.133|:443... connected. HTTP request sent, awaiting response... 200 OK Length: 116068 (113K) [image/jpeg] Saving to: ‘panda.jpg’ panda.jpg 100%[===================>] 113.35K --.-KB/s in 0.04s 2019-07-01 03:55:54 (3.01 MB/s) - ‘panda.jpg’ saved [116068/116068] 
# Import image transformations
from efficientnet import center_crop_and_resize, preprocess_input

image = plt.imread('panda.jpg')

plt.figure(figsize=(10, 10))
plt.imshow(image)
plt.show()
Image in a Jupyter notebook
# preprocess input
image_size = model.input_shape[1]
x = center_crop_and_resize(image, image_size=image_size)
x = preprocess_input(x)
x = np.expand_dims(x, 0)

# make prediction and decode
y = model.predict(x)
decode_predictions(y)
Downloading data from https://storage.googleapis.com/download.tensorflow.org/data/imagenet_class_index.json 40960/35363 [==================================] - 0s 0us/step
[[('n02510455', 'giant_panda', 0.8347929), ('n02134084', 'ice_bear', 0.0156020615), ('n02509815', 'lesser_panda', 0.004553518), ('n02133161', 'American_black_bear', 0.0024719085), ('n02132136', 'brown_bear', 0.002070759)]]

ResNet50 in Keras

# Import required modules
from keras.applications.resnet50 import ResNet50

model = ResNet50(weights='imagenet')
W0701 04:09:44.433862 139951651768192 deprecation_wrapper.py:119] From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:3976: The name tf.nn.max_pool is deprecated. Please use tf.nn.max_pool2d instead.
Downloading data from https://github.com/fchollet/deep-learning-models/releases/download/v0.2/resnet50_weights_tf_dim_ordering_tf_kernels.h5 102858752/102853048 [==============================] - 1s 0us/step 
# preprocess input
image_size = model.input_shape[1]
x = center_crop_and_resize(image, image_size=image_size)
x = preprocess_input(x)
x = np.expand_dims(x, 0)

# make prediction and decode
y = model.predict(x)
decode_predictions(y)
[[('n01930112', 'nematode', 0.029634755), ('n03729826', 'matchstick', 0.028414724), ('n03657121', 'lens_cap', 0.022990918), ('n04153751', 'screw', 0.022090413), ('n02783161', 'ballpoint', 0.021520559)]]

Now that we know how to run inference on both models, let's start off with our testing. For the testing, we will use time module to keep a track of time taken for inference. Since we are using the same module for image pre-processing, we will not take it into consideration and only stick to inference time. We will run the inference 1000 times and take an average of all the times.

import time

# First let's initialize our lists of inference times
efficientNetTimeInference = []
resNetTimeInference = []

# Preprocess image in advance
image_size = model.input_shape[1]
x = center_crop_and_resize(image, image_size=image_size)
x = preprocess_input(x)
x = np.expand_dims(x, 0)

resNet = ResNet50(weights='imagenet')
efficientNet = EfficientNetB0(weights='imagenet')

for i in range(100):
  print("Iteration Number: {}".format(str(i).zfill(3)))
  startTime = time.time()
  # First, EfficientNet-B0
  y = efficientNet.predict(x)
  efficientNetTimeInference.append(time.time()-startTime)
  # Next, ResNet50
  startTime = time.time()
  y = resNet.predict(x)
  resNetTimeInference.append(time.time()-startTime)
Iteration Number: 000 Iteration Number: 001 Iteration Number: 002 Iteration Number: 003 Iteration Number: 004 Iteration Number: 005 Iteration Number: 006 Iteration Number: 007 Iteration Number: 008 Iteration Number: 009 Iteration Number: 010 Iteration Number: 011 Iteration Number: 012 Iteration Number: 013 Iteration Number: 014 Iteration Number: 015 Iteration Number: 016 Iteration Number: 017 Iteration Number: 018 Iteration Number: 019 Iteration Number: 020 Iteration Number: 021 Iteration Number: 022 Iteration Number: 023 Iteration Number: 024 Iteration Number: 025 Iteration Number: 026 Iteration Number: 027 Iteration Number: 028 Iteration Number: 029 Iteration Number: 030 Iteration Number: 031 Iteration Number: 032 Iteration Number: 033 Iteration Number: 034 Iteration Number: 035 Iteration Number: 036 Iteration Number: 037 Iteration Number: 038 Iteration Number: 039 Iteration Number: 040 Iteration Number: 041 Iteration Number: 042 Iteration Number: 043 Iteration Number: 044 Iteration Number: 045 Iteration Number: 046 Iteration Number: 047 Iteration Number: 048 Iteration Number: 049 Iteration Number: 050 Iteration Number: 051 Iteration Number: 052 Iteration Number: 053 Iteration Number: 054 Iteration Number: 055 Iteration Number: 056 Iteration Number: 057 Iteration Number: 058 Iteration Number: 059 Iteration Number: 060 Iteration Number: 061 Iteration Number: 062 Iteration Number: 063 Iteration Number: 064 Iteration Number: 065 Iteration Number: 066 Iteration Number: 067 Iteration Number: 068 Iteration Number: 069 Iteration Number: 070 Iteration Number: 071 Iteration Number: 072 Iteration Number: 073 Iteration Number: 074 Iteration Number: 075 Iteration Number: 076 Iteration Number: 077 Iteration Number: 078 Iteration Number: 079 Iteration Number: 080 Iteration Number: 081 Iteration Number: 082 Iteration Number: 083 Iteration Number: 084 Iteration Number: 085 Iteration Number: 086 Iteration Number: 087 Iteration Number: 088 Iteration Number: 089 Iteration Number: 090 Iteration Number: 091 Iteration Number: 092 Iteration Number: 093 Iteration Number: 094 Iteration Number: 095 Iteration Number: 096 Iteration Number: 097 Iteration Number: 098 Iteration Number: 099 
# Print average inference time
print("Average inference time for EfficientNet-B0: {}".format(np.mean(efficientNetTimeInference)))
print("Average inference time for ResNet50: {}".format(np.mean(resNetTimeInference)))
Average inference time for EfficientNet-B0: 0.1576129364967346 Average inference time for ResNet50: 0.30902464866638185

EfficientNet Inference in PyTorch

# Load required modules
import json
from PIL import Image

import torch
from torchvision import transforms

# Load model from torch hub
model = torch.hub.load('rwightman/gen-efficientnet-pytorch', 'efficientnet_b0', pretrained=True)
Downloading: "https://github.com/rwightman/gen-efficientnet-pytorch/archive/master.zip" to /root/.cache/torch/hub/master.zip Downloading: "https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/efficientnet_b0-d6904d92.pth" to /root/.cache/torch/checkpoints/efficientnet_b0-d6904d92.pth 100%|██████████| 21376958/21376958 [00:01<00:00, 18063640.04it/s] 
img = Image.open('panda.jpg')
img
Image in a Jupyter notebook
# Preprocess image
tfms = transforms.Compose([transforms.Resize(image_size), transforms.CenterCrop(image_size), 
                           transforms.ToTensor(),
                           transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),])
img = tfms(img).unsqueeze(0)

# Download labels file
!wget https://raw.githubusercontent.com/lukemelas/EfficientNet-PyTorch/master/examples/simple/labels_map.txt
--2019-07-01 04:14:06-- https://raw.githubusercontent.com/lukemelas/EfficientNet-PyTorch/master/examples/simple/labels_map.txt Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.0.133, 151.101.64.133, 151.101.128.133, ... Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.0.133|:443... connected. HTTP request sent, awaiting response... 200 OK Length: 31565 (31K) [text/plain] Saving to: ‘labels_map.txt’ labels_map.txt 100%[===================>] 30.83K --.-KB/s in 0.01s 2019-07-01 04:14:06 (2.46 MB/s) - ‘labels_map.txt’ saved [31565/31565] 
# Load class names
labels_map = json.load(open('labels_map.txt'))
labels_map = [labels_map[str(i)] for i in range(1000)]

model.eval()
with torch.no_grad():
    logits = model(img)
preds = torch.topk(logits, k=5).indices.squeeze(0).tolist()

for idx in preds:
    label = labels_map[idx]
    prob = torch.softmax(logits, dim=1)[0, idx].item()
    print('{:<75} ({:.2f}%)'.format(label, prob*100))
giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca (90.13%) ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus (0.52%) lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens (0.32%) American black bear, black bear, Ursus americanus, Euarctos americanus (0.10%) soccer ball (0.08%)

ResNet50 Inference in PyTorch

# Load required modules
import torchvision.models as models
# Load pretrained model
resnet = models.resnet50(pretrained=True)

resnet.eval()
with torch.no_grad():
    logits = resnet(img)
preds = torch.topk(logits, k=5).indices.squeeze(0).tolist()

for idx in preds:
    label = labels_map[idx]
    prob = torch.softmax(logits, dim=1)[0, idx].item()
    print('{:<75} ({:.2f}%)'.format(label, prob*100))
giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca (99.38%) French bulldog (0.09%) hog, pig, grunter, squealer, Sus scrofa (0.04%) ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus (0.03%) soccer ball (0.03%)

Now, as before, let's run a speed test to see how much time the inference takes for both the models.

# First let's initialize our lists of inference times
efficientNetTimeInference = []
resNetTimeInference = []

resNet = models.resnet50(pretrained=True)
efficientNet = torch.hub.load('rwightman/gen-efficientnet-pytorch', 'efficientnet_b0', pretrained=True)
Using cache found in /root/.cache/torch/hub/rwightman_gen-efficientnet-pytorch_master 
for i in range(100):
  print("Iteration Number: {}".format(str(i).zfill(3)))
  startTime = time.time()
  # First, EfficientNet-B0
  efficientNet.eval()
  with torch.no_grad():
    logits = efficientNet(img)
  efficientNetTimeInference.append(time.time()-startTime)
  # Next, MobileNet
  startTime = time.time()
  resNet.eval()
  with torch.no_grad():
    logits = resNet(img)
  resNetTimeInference.append(time.time()-startTime)
Iteration Number: 000 Iteration Number: 001 Iteration Number: 002 Iteration Number: 003 Iteration Number: 004 Iteration Number: 005 Iteration Number: 006 Iteration Number: 007 Iteration Number: 008 Iteration Number: 009 Iteration Number: 010 Iteration Number: 011 Iteration Number: 012 Iteration Number: 013 Iteration Number: 014 Iteration Number: 015 Iteration Number: 016 Iteration Number: 017 Iteration Number: 018 Iteration Number: 019 Iteration Number: 020 Iteration Number: 021 Iteration Number: 022 Iteration Number: 023 Iteration Number: 024 Iteration Number: 025 Iteration Number: 026 Iteration Number: 027 Iteration Number: 028 Iteration Number: 029 Iteration Number: 030 Iteration Number: 031 Iteration Number: 032 Iteration Number: 033 Iteration Number: 034 Iteration Number: 035 Iteration Number: 036 Iteration Number: 037 Iteration Number: 038 Iteration Number: 039 Iteration Number: 040 Iteration Number: 041 Iteration Number: 042 Iteration Number: 043 Iteration Number: 044 Iteration Number: 045 Iteration Number: 046 Iteration Number: 047 Iteration Number: 048 Iteration Number: 049 Iteration Number: 050 Iteration Number: 051 Iteration Number: 052 Iteration Number: 053 Iteration Number: 054 Iteration Number: 055 Iteration Number: 056 Iteration Number: 057 Iteration Number: 058 Iteration Number: 059 Iteration Number: 060 Iteration Number: 061 Iteration Number: 062 Iteration Number: 063 Iteration Number: 064 Iteration Number: 065 Iteration Number: 066 Iteration Number: 067 Iteration Number: 068 Iteration Number: 069 Iteration Number: 070 Iteration Number: 071 Iteration Number: 072 Iteration Number: 073 Iteration Number: 074 Iteration Number: 075 Iteration Number: 076 Iteration Number: 077 Iteration Number: 078 Iteration Number: 079 Iteration Number: 080 Iteration Number: 081 Iteration Number: 082 Iteration Number: 083 Iteration Number: 084 Iteration Number: 085 Iteration Number: 086 Iteration Number: 087 Iteration Number: 088 Iteration Number: 089 Iteration Number: 090 Iteration Number: 091 Iteration Number: 092 Iteration Number: 093 Iteration Number: 094 Iteration Number: 095 Iteration Number: 096 Iteration Number: 097 Iteration Number: 098 Iteration Number: 099 
# Print average inference time
print("Average inference time for EfficientNet-B0: {}".format(np.mean(efficientNetTimeInference)))
print("Average inference time for ResNet50: {}".format(np.mean(resNetTimeInference)))
Average inference time for EfficientNet-B0: 0.12161755561828613 Average inference time for ResNet50: 0.28322251319885255