jamesonthecrow

guard let pose = poseResult.decodePose() else { return }

let leftArmParts: [PosePart] = [.leftWrist, .leftElbow, .leftShoulder]
let rightArmParts: [PosePart] = [.rightWrist, .rightElbow, .rightShoulder]

var foundLeftArm: [Keypoint] = []
var foundRightArm: [Keypoint] = []

for keypoint in pose.keypoints {
    if leftArmParts.contains(keypoint.part) {

jamesonthecrow

# A script to train an artistic style transfer model from a custom style image.
# A Google Colab going through the same steps can be found here:
# https://colab.research.google.com/drive/1nDkxLKBgZGFscGoF0tfyPMGqW03xITl0#scrollTo=V33xVH-CWUCs
# Note that this script will download and unzip 1GB of photos for training.
# Make sure you have the appropriate permissions to use any images.

# CHANGE ME BEFORE RUNNING
STYLE_IMAGE_URL='STYLE_IMAGE_URL'

# Install requirements

jamesonthecrow

import Fritz

class PoseEstimationViewController: UIViewController {

    private let poseModel = FritzVisionPoseModel()
    // We can also set of sensitivity parameters for our model.
    // The poseThreshold is a number between 0 and 1. Higher numbers mean
    // the model must be more confident about its estimate, thus reducing false
    // positives.
    internal var poseThreshold: Double = 0.3

jamesonthecrow

import ai.fritz.vision.FritzVision;
import ai.fritz.poseestimationmodel.PoseEstimationOnDeviceModel;
import ai.fritz.vision.poseestimation.FritzVisionPosePredictor;
import ai.fritz.core.FritzOnDeviceModel;
// ...

public class CameraActivity extends Activity implements ImageReader.OnImageAvailableListener {

    private FritzVisionPosePredictor posePredictor;
    private FritzVisionPoseResult poseResult;

jamesonthecrow

import CreateML
import Foundation

// Load our data into an MLDataTable object.
let dataFilename = "PATH/TO/data.json"
let data = try MLDataTable(contentsOf: URL(fileURLWithPath: dataFilename))
print(data.description)
/*
Columns:
    label	string

jamesonthecrow

extension PoseEstimationViewController: AVCaptureVideoDataOutputSampleBufferDelegate {

    func captureOutput(_ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from connection: AVCaptureConnection) {
        // FritzVisionImage objects offer convient ways to manipulate
        // images used as input to machine learning models.
        // You can resize, crop, and scale images to your needs.
        let image = FritzVisionImage(sampleBuffer: sampleBuffer, connection: connection)

        // Set options for our pose estimation model using the constants
        // we initialized earlier in the ViewController.

jamesonthecrow

class ViewController: UIViewController, UIImagePickerControllerDelegate,
UINavigationControllerDelegate {
  /// The rest of the view controller...
  
  /// Scores output from model greater than this value will be set as 1.
  /// Lowering this value will make the mask more intense for lower confidence values.
  var clippingScoresAbove: Double { return 0.6 }
  
  /// Values lower than this value will not appear in the mask.
  var zeroingScoresBelow: Double { return 0.4 }

jamesonthecrow

// Initialize the model included with the app
PetSegmentationOnDeviceModel onDeviceModel = new PetSegmentationOnDeviceModel();

FritzVisionSegmentPredictorOptions options = new FritzVisionSegmentPredictorOptions.Builder()
  .targetConfidenceThreshold(.4f)
  .build();

// Create the predictor with the Pet Segmentation model.
predictor = FritzVision.ImageSegmentation.getPredictor(onDeviceModel, options);

jamesonthecrow

keras.backend.clear_session()
# Retrain the model with our new configuration and callback
model = build_model()

model.compile(
    keras.optimizers.Adam(lr=metadata['learning_rate']),
    loss=keras.losses.sparse_categorical_crossentropy,
    metrics=[keras.metrics.sparse_categorical_accuracy]
)

jamesonthecrow

import fritz
import fritz.train

# Fritz needs to be configured first. Calling the fritz.Configure() method will
# read the credentials we setup for the CLI earlier.
fritz.configure()

# Create the callback

# Start by defining a training configuration and storing it as metadata