This README is the one indexed from the repository at its latest ref, not from the tag for this version.

Tensorflow Lite image tools

a library for image classification and feature detection with tflite and crystal lang

Installation

1. Add the dependency to your shard.yml:

   dependencies:
     tflite_image:
       github: spider-gazelle/tflite_image

2. Run shards install

Usage

Image classification

require "tflite_image"

# init the tensorflow client with your classification model
client = TensorflowLite::Client.new("./models/classifier.tflite")

# init the classifier
classifier = TensorflowLite::Image::Classification.new(client)

# load your image
canvas = StumpyJPEG.read("./some_image.jpg")

# run the model, outputs the scaled image that was run through the model
scaled_canvas, detections = classifier.run canvas

# parse the outputs
puts detections.inspect

Object detection

require "tflite_image"

# init the tensorflow client with your object detection model
client = TensorflowLite::Client.new("./models/detector.tflite")

# init the detector
detector = TensorflowLite::Image::ObjectDetection.new(client, scale_mode: :cover)

# load your image
canvas = StumpyJPEG.read("./some_image.jpg")

# run the model, outputs the scaled image that was run through the model
scaled_canvas, detections = detector.run canvas

# parse the outputs
puts detections.inspect

# markup the image with bounding boxes and save the output:
# ========================================================

# we need to apply offsets to the detections
# as they apply to the scaled_canvas
# so they need adjustment to be mapped back onto the original image
offsets = detector.detection_adjustments(canvas)
detector.markup canvas, detections, *offsets
StumpyPNG.write(canvas, "./bin/detection_output.png")

Pose detection

require "tflite_image"

# init the tensorflow client with your object detection model
client = TensorflowLite::Client.new("./models/pose.tflite")

# init the detector
pose = TensorflowLite::Image::PoseEstimation.new(client)

# load your image
canvas = StumpyJPEG.read("./person_image.jpg")

# run the model, outputs the scaled image that was run through the model
scaled_canvas, detections = pose.run canvas

# parse the outputs
puts detections.inspect

# markup the image with a skeleton and save the output:
# ========================================================

# we need to apply offsets to the detections
# as they apply to the scaled_canvas
# so they need adjustment to be mapped back onto the original image
offsets = pose.detection_adjustments(canvas)
pose.markup canvas, detections, *offsets
StumpyPNG.write(canvas, "./bin/pose_output.png")

Image Segmentation

require "tflite_image"

# init the tensorflow client with your object detection model
client = TensorflowLite::Client.new(
  model: URI.parse("https://raw.githubusercontent.com/google-coral/test_data/master/deeplabv3_mnv2_pascal_quant.tflite"),
  labels: URI.parse("https://raw.githubusercontent.com/google-coral/test_data/master/pascal_voc_segmentation_labels.txt")
)

# init the segmentation
seg = Segmentation.new(client)

# load your image
canvas = StumpyJPEG.read("./image.jpg")

# run the model, outputs the scaled image that was run through the model
scaled_canvas, detections = seg.run canvas
scaled_mask = seg.build_image_mask detections

# parse the outputs
puts "Unique objects found! pixels: #{detections.pixels.size}, unique: #{detections.labels_detected}"

# scale the mask to match the input image
# ========================================================

# create a mask for the original image
offsets = pose.detection_adjustments(canvas)
mask = seg.scale_image_mask(canvas, scaled_mask, *offsets)
StumpyPNG.write(mask, "./bin/mask.png")

# overlay the mask onto the original image
canvas.paste(mask, 0, 0)
StumpyPNG.write(canvas, "./bin/masked_image.png")

Contributing

1. Fork it (https://github.com/spider-gazelle/tflite_image/fork)
2. Create your feature branch (git checkout -b my-new-feature)
3. Commit your changes (git commit -am 'Add some feature')
4. Push to the branch (git push origin my-new-feature)
5. Create a new Pull Request

Contributors

• Stephen von Takach - creator and maintainer