Classification models Zoo - Keras (and TensorFlow Keras)

Trained on ImageNet classification models. The library is designed to work both with Keras and TensorFlow Keras. See example below.

Important!

There was a huge library update 05 of August. Now classification-models works with both frameworks: keras and tensorflow.keras. If you have models, trained before that date, to load them, please, use image-classifiers (PyPI package name) of 0.2.2 version. You can roll back using pip install -U image-classifiers==0.2.2.

Architectures:

• VGG [16, 19]
• ResNet [18, 34, 50, 101, 152]
• ResNeXt [50, 101]
• SE-ResNet [18, 34, 50, 101, 152]
• SE-ResNeXt [50, 101]
• SE-Net [154]
• DenseNet [121, 169, 201]
• Inception ResNet V2
• Inception V3
• Xception
• NASNet [large, mobile]
• MobileNet
• MobileNet v2

Specification

The top-k accuracy were obtained using center single crop on the 2012 ILSVRC ImageNet validation set and may differ from the original ones. The input size used was 224x224 (min size 256) for all models except:

• NASNetLarge 331x331 (352)
• InceptionV3 299x299 (324)
• InceptionResNetV2 299x299 (324)
• Xception 299x299 (324)

The inference *Time was evaluated on 500 batches of size 16. All models have been tested using same hardware and software. Time is listed just for comparison of performance.

Weights

Installation

Requirements:

• Keras >= 2.2.0 / TensorFlow >= 1.12
• keras_applications >= 1.0.7

Note

This library does not have TensorFlow in a requirements for installation. 
Please, choose suitable version (‘cpu’/’gpu’) and install it manually using 
official Guide (https://www.tensorflow.org/install/).

PyPI stable package:

$ pip install image-classifiers==0.2.2

PyPI latest package:

$ pip install image-classifiers==1.0.0b1

Latest version:

$ pip install git+https://github.com/qubvel/classification_models.git

Examples

Loading model with imagenet weights:

# for keras
from classification_models.keras import Classifiers

# for tensorflow.keras
# from classification_models.tfkeras import Classifiers

ResNet18, preprocess_input = Classifiers.get('resnet18')
model = ResNet18((224, 224, 3), weights='imagenet')

This way take one additional line of code, however if you would like to train several models you do not need to import them directly, just access everything through Classifiers.

You can get all model names using Classifiers.models_names() method.

Inference example:

import numpy as np
from skimage.io import imread
from skimage.transform import resize
from keras.applications.imagenet_utils import decode_predictions
from classification_models.keras import Classifiers

ResNet18, preprocess_input = Classifiers.get('resnet18')

# read and prepare image
x = imread('./imgs/tests/seagull.jpg')
x = resize(x, (224, 224)) * 255    # cast back to 0-255 range
x = preprocess_input(x)
x = np.expand_dims(x, 0)

# load model
model = ResNet18(input_shape=(224,224,3), weights='imagenet', classes=1000)

# processing image
y = model.predict(x)

# result
print(decode_predictions(y))

Model fine-tuning example:

import keras
from classification_models.keras import Classifiers

ResNet18, preprocess_input = Classifiers.get('resnet18')

# prepare your data
X = ...
y = ...

X = preprocess_input(X)

n_classes = 10

# build model
base_model = ResNet18(input_shape=(224,224,3), weights='imagenet', include_top=False)
x = keras.layers.GlobalAveragePooling2D()(base_model.output)
output = keras.layers.Dense(n_classes, activation='softmax')(x)
model = keras.models.Model(inputs=[base_model.input], outputs=[output])

# train
model.compile(optimizer='SGD', loss='categorical_crossentropy', metrics=['accuracy'])
model.fit(X, y)