Large-scale image classification models on Keras

This is a collection of large-scale image classification models. Many of them are pretrained on ImageNet-1K dataset and loaded automatically during use. All pretrained models require the same ordinary normalization. Scripts for training/evaluating/converting models are in the imgclsmob repo.

List of implemented models

• AlexNet ('One weird trick for parallelizing convolutional neural networks')
• VGG/BN-VGG ('Very Deep Convolutional Networks for Large-Scale Image Recognition')
• ResNet ('Deep Residual Learning for Image Recognition')
• PreResNet ('Identity Mappings in Deep Residual Networks')
• ResNeXt ('Aggregated Residual Transformations for Deep Neural Networks')
• SENet/SE-ResNet/SE-PreResNet/SE-ResNeXt ('Squeeze-and-Excitation Networks')
• DenseNet ('Densely Connected Convolutional Networks')
• DarkNet Ref/Tiny/19 ('Darknet: Open source neural networks in c')
• DarkNet-53 ('YOLOv3: An Incremental Improvement')
• SqueezeNet/SqueezeResNet ('SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size')
• SqueezeNext ('SqueezeNext: Hardware-Aware Neural Network Design')
• ShuffleNet ('ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices')
• ShuffleNetV2 ('ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design')
• MENet ('Merging and Evolution: Improving Convolutional Neural Networks for Mobile Applications')
• MobileNet ('MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications')
• FD-MobileNet ('FD-MobileNet: Improved MobileNet with A Fast Downsampling Strategy')
• MobileNetV2 ('MobileNetV2: Inverted Residuals and Linear Bottlenecks')
• IGCV3 ('IGCV3: Interleaved Low-Rank Group Convolutions for Efficient Deep Neural Networks')
• MnasNet ('MnasNet: Platform-Aware Neural Architecture Search for Mobile')
• EfficientNet ('EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks')

Installation

To use the models in your project, simply install the kerascv package with desired backend. For example for MXNet backend:

pip install mxnet>=1.2.1 keras-mxnet kerascv

Or if you prefer TensorFlow backend:

pip install tensorflow kerascv

To enable/disable different hardware supports, check out installation instruction for the corresponding backend.

After installation check that the backend field is set to the correct value in the file ~/.keras/keras.json. It is also preferable to set the value of the image_data_format field to channels_first in the case of using the MXNet backend.

Usage

Example of using a pretrained ResNet-18 model (for channels_first data format):

from kerascv.model_provider import get_model as kecv_get_model
import numpy as np

net = kecv_get_model("resnet18", pretrained=True)
x = np.zeros((1, 3, 224, 224), np.float32)
y = net.predict(x)

Pretrained models (ImageNet-1K)

Some remarks:

• All quality values are estimated with MXNet backend.
• Top1/Top5 are the standard 1-crop Top-1/Top-5 errors (in percents) on the validation subset of the ImageNet-1K dataset.
• FLOPs/2 is the number of FLOPs divided by two to be similar to the number of MACs.
• Remark Converted from GL model means that the model was trained on MXNet/Gluon and then converted to Keras.