Things (objects, people, animals) detection using DJL / published by https://github.com/dacr/code-examples-manager #55ebb071-12a6-4268-b6b2-2a6d835eddab/82dadeb130fd3ebeb1a823871ea5e71fb79a992

djl-object-detection.sc

// summary : Things (objects, people, animals) detection using DJL
// keywords : djl, machine-learning, tutorial, detection, ai, @testable
// publish : gist
// authors : David Crosson
// license : Apache NON-AI License Version 2.0 (https://raw.githubusercontent.com/non-ai-licenses/non-ai-licenses/main/NON-AI-APACHE2)
// id : 55ebb071-12a6-4268-b6b2-2a6d835eddab
// created-on : 2021-03-05T17:40:29Z
// managed-by : https://github.com/dacr/code-examples-manager
// run-with : scala-cli $file

// ---------------------
//> using scala "3.4.2"
//> using dep "org.slf4j:slf4j-api:2.0.13"
//> using dep "org.slf4j:slf4j-simple:2.0.13"
//> using dep "net.java.dev.jna:jna:5.14.0"

//> using dep "ai.djl:api:0.29.0"
//> using dep "ai.djl:basicdataset:0.29.0"
//> using dep "ai.djl:model-zoo:0.29.0"
//> using dep "ai.djl.huggingface:tokenizers:0.29.0"

//> using dep "ai.djl.mxnet:mxnet-engine:0.29.0"
//> using dep "ai.djl.mxnet:mxnet-model-zoo:0.29.0"

//> using dep "ai.djl.pytorch:pytorch-engine:0.29.0"
//> using dep "ai.djl.pytorch:pytorch-model-zoo:0.29.0"

//> using dep "ai.djl.tensorflow:tensorflow-engine:0.29.0"
//> using dep "ai.djl.tensorflow:tensorflow-model-zoo:0.29.0"

//> using dep "ai.djl.onnxruntime:onnxruntime-engine:0.29.0"

// ---------------------

// inspired from https://docs.djl.ai/examples/docs/object_detection.html

//System.setProperty("org.slf4j.simpleLogger.defaultLogLevel", "debug")

import ai.djl.Application
import ai.djl.engine.Engine
import ai.djl.modality.cv.Image
import ai.djl.modality.cv.ImageFactory
import ai.djl.modality.cv.output.DetectedObjects
import ai.djl.modality.cv.output.DetectedObjects.DetectedObject
import ai.djl.modality.cv.translator.YoloV8TranslatorFactory
import ai.djl.repository.zoo.Criteria
import ai.djl.repository.zoo.ModelZoo
import ai.djl.repository.zoo.ZooModel
import ai.djl.training.util.ProgressBar

import java.nio.file.Files
import java.nio.file.Path
import java.nio.file.Paths
import scala.jdk.CollectionConverters.*

// ----------------------------------------------------------------------------------------------

def saveBoundingBoxImage(img: Image, detection: DetectedObjects, outputFile: Path): Unit = {
  val newImage = img.duplicate()
  newImage.drawBoundingBoxes(detection)
  import java.nio.file.Files
  newImage.save(Files.newOutputStream(outputFile), "png")
}

def basename(filename: String): String = {
  filename
    .split("[/](?=[^/]*$)", 2)
    .last
    .split("[.]", 2)
    .head
}

// ----------------------------------------------------------------------------------------------
val inputImageURL   = "https://mapland.fr/data/ai/images-samples/example-016.jpg"
val outputDir       = Paths.get("build/output")
Files.createDirectories(outputDir)
val outputImageFile = outputDir.resolve("detected-objects-" + basename(inputImageURL) + ".png")
// ----------------------------------------------------------------------------------------------

//val engineName = Engine.getDefaultEngineName()
//val engineName = "TensorFlow"
//val engineName = "PyTorch"
//println(s"Using engine name :  $engineName (default is ${Engine.getDefaultEngineName()})")
val criteria =
  Criteria.builder
    .setTypes(classOf[Image], classOf[DetectedObjects])
    .optModelUrls("djl://ai.djl.onnxruntime/yolov8n")
    .optEngine("OnnxRuntime")
    //.optApplication(Application.CV.OBJECT_DETECTION)
    // .optFilters(Map("backbone" -> "mobilenet1.0", "imageSize" -> "416", "dataset" -> "coco").asJava)
    // .optFilter("backbone", "resnet50")
    // .optFilter("backbone", "vgg16")
    //.optFilters(Map("backbone" -> "darknet53", "imageSize" -> "416", "dataset" -> "coco").asJava)
    // .optFilter("backbone", "mobilenet_v2") // TensorFlow
    // .optEngine(engineName)
    // .optEngine("OnnxRuntime")
    .optArgument("width", 640)
    .optArgument("height", 640)
    .optArgument("resize", true)
    .optArgument("toTensor", true)
    .optArgument("applyRatio", true)
    .optArgument("threshold", 0.4f)
    .optArgument("maxBox", 2000)
    .optTranslatorFactory(new YoloV8TranslatorFactory())
    .optProgress(new ProgressBar)
    .build

val model = ModelZoo.loadModel(criteria)
println(s"Using ${model.getName} ${model.getModelPath}")

val predictor = model.newPredictor()

val img = ImageFactory.getInstance().fromUrl(inputImageURL)

val detection: DetectedObjects = predictor.predict(img)
println("number of detected object : " + detection.getNumberOfObjects)
detection
  .items()
  .asScala
  .toList
  .asInstanceOf[List[DetectedObject]]
  .sortBy(_.getProbability)
  // .filter(_.getProbability > 0.4d)
  .foreach { ob =>
    println(ob.getClassName + " " + ob.getProbability)
  }
saveBoundingBoxImage(img, detection, outputImageFile)
println("images with detected things bounding box saves as " + outputImageFile)