https://github.com/titu1994/Keras-NASNet

"NASNet" models in Keras 2.0+ with weights
https://github.com/titu1994/Keras-NASNet

keras nasnet-models tensorflow

Last synced: 3 days ago
JSON representation

"NASNet" models in Keras 2.0+ with weights

• Host: GitHub
• URL: https://github.com/titu1994/Keras-NASNet
• Owner: titu1994
• License: mit
• Created: 2017-11-17T03:50:57.000Z (almost 7 years ago)
• Default Branch: master
• Last Pushed: 2018-05-17T06:22:34.000Z (over 6 years ago)
• Last Synced: 2024-10-27T12:31:03.879Z (7 days ago)
• Topics: keras, nasnet-models, tensorflow
• Language: Python
• Homepage:
• Size: 3.55 MB
• Stars: 200
• Watchers: 20
• Forks: 64
• Open Issues: 9
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-image-classification - unofficial-keras : https://github.com/titu1994/Keras-NASNet
• awesome-image-classification - unofficial-keras : https://github.com/titu1994/Keras-NASNet

README

        
# Keras Neural Architecture Search Network (NASNet)

An implementation of "NASNet" models from the paper [Learning Transferable Architectures for Scalable Image Recognitio](https://arxiv.org/abs/1707.07012) in Keras 2.0+.

Based on the models described in the [TFSlim implementation](https://github.com/tensorflow/models/tree/master/research/slim/nets/nasnet) and some modules from the [TensorNets implementation](https://github.com/taehoonlee/tensornets/blob/master/tensornets/nasnets.py)

Weights have been ported over from the official [NASNet Tensorflow repository](https://github.com/tensorflow/models/tree/master/research/slim/nets/nasnet).

Since CIFAR weights are not provided, and I don't have the resources to train such large models on CIFAR, those weights will not be provided. External help is appreciated.

# Usage

All types of NASNet models can be built. In addition, `NASNet Large - NASNet (6 @ 4032)` and `NASNet Mobile - NASNet (4 @ 1056)` are prebuilt and provided as `NASNetLarge` and `NASNetMobile`.

## Building a speficific NASNet model

```python

from nasnet import NASNet

# the parameters for NASNetLarge

model = NASNet(input_shape=(331, 331, 3),

           penultimate_filters=4032,

           nb_blocks=6,

           stem_filters=96,

           skip_reduction=True,

           use_auxilary_branch=False,

           filters_multiplier=2,

           dropout=0.5,

           classes=1000)

```

## Using Pre-built NASNet models

```python

from nasnet import NASNetLarge, NASNetMobile

model = NASNetLarge(input_shape=(331, 331, 3), dropout=0.5)

```

# Network Overview