https://github.com/en10/kerasinception

Google Inception-V3 with Keras
https://github.com/en10/kerasinception

image-classification image-recognition inception-v3 inceptionv3 keras keras-tensorflow tensorflow

Last synced: 4 months ago
JSON representation

Google Inception-V3 with Keras

• Host: GitHub
• URL: https://github.com/en10/kerasinception
• Owner: EN10
• Created: 2017-05-20T13:45:33.000Z (about 9 years ago)
• Default Branch: master
• Last Pushed: 2018-02-17T18:44:49.000Z (over 8 years ago)
• Last Synced: 2025-08-01T04:45:20.548Z (11 months ago)
• Topics: image-classification, image-recognition, inception-v3, inceptionv3, keras, keras-tensorflow, tensorflow
• Language: Python
• Homepage:
• Size: 886 KB
• Stars: 11
• Watchers: 3
• Forks: 8
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Inception Keras

#### Image Recognition using Keras and Inception-v3

Keras allows ['easy and fast'](https://keras.io) use of models: [example](https://keras.io/applications/#usage-examples-for-image-classification-models).  

[Inception-v3](https://www.tensorflow.org/tutorials/image_recognition) is a trained image recognition model for tensorflow with 93.9% Top-5 Accuracy.     

* **[Other Keras Models](https://github.com/EN10/KerasModels)**

![Inception arch](https://github.com/EN10/inception-keras/raw/master/images/Inception-v3.png)

It was trained on 1.2 million images from [ImageNet](http://image-net.org/challenges/LSVRC/2012/index#introduction) to detect [1000 classes](http://image-net.org/challenges/LSVRC/2012/browse-synsets) (or labels for the images).  

![ImageNet](https://github.com/EN10/inception-keras/raw/master/images/ImageNet.png)

It took [8 NVIDIA Tesla K40s, 2 weeks](https://research.googleblog.com/2016/03/train-your-own-image-classifier-with.html) to train.

![K40](https://github.com/EN10/inception-keras/raw/master/images/K40.jpg)

This example was built and tested on c9.io or cs50.io as they provide a free Ubuntu VM (docker container) with 512MB RAM and 2GB Disk.

## Install

    sudo pip install -U pip 

    sudo pip install tensorflow 

    sudo pip install h5py pillow 

`sudo pip install -U pip` update pip with tensorflow  

`... h5py` to load weights  

`... pillow` to load image  

as `from tensorflow.python.keras ... import` is used, as `keras` is not installed separately  

`keras` path correct as of `tensorflow 1.4.1`

## Run

`python predict.py`  

Line 7: Loads image `'image.jpg'` then runs inference on `InceptionV3`.  

Line 14: Outputs top 5 predictions and probabilities.   

### Performance

**To disable `"cpu_feature_guard"`:**  

`export TF_CPP_MIN_LOG_LEVEL=2`  

Improve Performance:    

`TensorFlow binary compiled to use: SSE4.1 SSE4.2 AVX`

* [Precompiled](https://github.com/EN10/TensorFlow-For-Poets#performance)

* [FloydHub](https://github.com/EN10/FloydHub)

### Tensorflow Imports

[image](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/keras/python/keras/preprocessing/image.py)  

[inception_v3](https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/keras/python/keras/applications/inception_v3.py)  

[preprocess_input(x)](https://github.com/fchollet/keras/blob/master/keras/applications/imagenet_utils.py)