Skip to content

Latest commit

 

History

History
53 lines (40 loc) · 3.17 KB

README.adoc

File metadata and controls

53 lines (40 loc) · 3.17 KB
Raw

Image Recognition Processor

A processor that uses an Inception model to classify in real-time images into different categories (e.g. labels).

Model implements a deep Convolutional Neural Network that can achieve reasonable performance on hard visual recognition tasks - matching or exceeding human performance in some domains like image recognition.

The input of the model is an image as binary array.

The output is a JSON message in this format:

{
  "labels" : [
     {"giant panda":0.98649305}
  ]
}

Result contains the name of the recognized category (e.g. label) along with the confidence (e.g. confidence) that the image represents this category.

If the response-seize is set to value higher then 1, then the result will include the top response-seize probable labels. For example response-size=3 would return:

{
  "labels": [
    {"giant panda":0.98649305},
    {"badger":0.010562794},
    {"ice bear":0.001130851}
  ]
}

Payload

If the incoming type is byte[] and the content type is set to application/octet-stream , then the application process the input byte[] image into and outputs augmented byte[] image payload and json header.

Options

image.recognition.cache-model

cache the pre-trained tensorflow model. (Boolean, default: true)

image.recognition.debug-output

<documentation missing> (Boolean, default: false)

image.recognition.debug-output-path

<documentation missing> (String, default: image-recognition-result.png)

image.recognition.model

pre-trained tensorflow image recognition model. Note that the model must match the selected model type! (String, default: https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.4_224.tgz#mobilenet_v2_1.4_224_frozen.pb)

image.recognition.model-type

Supports three different pre-trained tensorflow image recognition models: Inception, MobileNetV1 and MobileNetV2 1. Inception graph uses 'input' as input and 'output' as output. 2. MobileNetV2 pre-trained models: https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet#pretrained-models - normalized image size is always square (e.g. H=W) - graph uses 'input' as input and 'MobilenetV2/Predictions/Reshape_1' as output. 3. MobileNetV1 pre-trained models: https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet_v1.md#pre-trained-models - graph uses 'input' as input and 'MobilenetV1/Predictions/Reshape_1' as output. (ModelType, default: <none>, possible values: inception,mobilenetv1,mobilenetv2)

image.recognition.normalized-image-size

Normalized image size. (Integer, default: 224)

image.recognition.response-size

number of recognized images. (Integer, default: 5)