deeplearning

deeplearning / InceptionNet / 1.0.5

README.md

Grace Hopper

Table of Contents

Introduction

This algorithm is a direct implementation of Google's InceptionNet, which was trained on the ImageNet 2015 dataset. It is implemented using Google's Tensorflow  python bindings,

the InceptionNet model architecture is as shown below:

googleNet

I/O

Input

source: String

  • source - (required) - the source image to classify, this must be in one of the following formats: data connector URI, web http/https resource file http://.. or https://..., or a base64 encoded JPEG String.

Output

{  
   "tags":[  
      {  
         "confidence": Double,
         "class": String
      },
      {  
         "confidence": Double,
         "class": String
      },
      {  
         "confidence": Double,
         "class": String
      },
      {  
         "confidence": Double,
         "class": String
      },
      {  
         "confidence":Double,
         "class": String
      }
   ]
}

  • tags - The top 5 classes that the model predicts are relevant to the image.
  • className - The classname for the class.
  • confidence - The confidence that this label is relevant.

Note: The first call to this algorithm will take a bit longer than sequential calls to due algorithm initialization. All following calls will be significantly faster.

Examples

Example 1

![car](http://i.imgur.com/ZiX34yx.jpg)

input

"http://i.imgur.com/LkrFjJP.jpg"

output

{  
   "tags":[  
      {  
         "class":"convertible",
         "confidence":0.3217212557792664
      },
      {  
         "class":"sports car, sport car",
         "confidence":0.18678018450737005
      },
      {  
         "class":"racer, race car, racing car",
         "confidence":0.09395135194063188
      },
      {  
         "class":"car wheel",
         "confidence":0.08665172755718234
      },
      {  
         "class":"grille, radiator grille",
         "confidence":0.07477507740259172
      }
   ]
}

Example 2

![dog](http://i.imgur.com/YKDmneL.jpg)

input

"http://i.imgur.com/YKDmneL.jpg"

output

{
  "tags":[
    {
      "class": "Samoyed, Samoyede",
      "confidence": 0.9004066586494446
    },
    {
      "class": "keeshond",
      "confidence": 0.004059927538037299
    },
    {
      "class": "Pomeranian",
      "confidence": 0.0024697233457118277
    },
    {
      "class": "Eskimo dog, husky",
      "confidence": 0.0013620780082419515
    },
    {
      "class": "Loafer",
      "confidence": 0.0011537882965058086
    }
  ]
}

Credits

For more info please check out Going Deeper With Convolutions by Christian Szegedy, Wei Liu et al

All sample images retrived from www.imgur.com on May 26, 2016.