Things (objects, people, animals) detection using DJL - compare models efficiency / published by https://github.com/dacr/code-examples-manager #bd813b80-9e47-489d-9a1d-86c5fb5c828e/215a493b6eb402293312d87dce57757676cdbfe

djl-object-detection-models-compare.sc

// summary : Things (objects, people, animals) detection using DJL - compare models efficiency
// 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 : bd813b80-9e47-489d-9a1d-86c5fb5c828e
// created-on : 2024-01-27T14:36:36+01:00
// 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"

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

System.setProperty("org.slf4j.simpleLogger.defaultLogLevel", "error")

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.repository.Artifact
import ai.djl.repository.zoo.{Criteria, ModelNotFoundException, ModelZoo, ModelZooResolver, ZooModel}
import ai.djl.training.util.ProgressBar

import java.net.{URI, URL}
import java.nio.file.Files
import java.nio.file.Path
import java.nio.file.Paths
import java.util.UUID
import java.util.concurrent.TimeUnit
import scala.concurrent.duration.Duration
import scala.jdk.CollectionConverters.*
import scala.io.AnsiColor.{BLUE, BOLD, CYAN, GREEN, MAGENTA, RED, RESET, UNDERLINED, YELLOW}
import scala.util.{Success, Try}

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

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 outputDir = Paths.get("build/output")
Files.createDirectories(outputDir)
// ----------------------------------------------------------------------------------------------

case class ModelArtifact(artifact: Artifact) {
  val uuid = UUID.nameUUIDFromBytes(
    s"$groupId$artifactId$version${properties.toList.sorted}".getBytes
  )

  def groupId: String                 = artifact.getMetadata.getGroupId
  def artifactId: String              = artifact.getMetadata.getArtifactId
  def version: String                 = artifact.getVersion
  def properties: Map[String, String] = artifact.getProperties.asScala.toMap

  def ident = toString()

  override def toString: String = s"$groupId:$artifactId:$version"
}

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

case class ModelResult(
  inputImageSource: URL,
  modelArtifact: ModelArtifact,
  selectedModelPath: Path,
  responseTime: Duration,
  detectedObjects: List[DetectedObject],
  generatedBoundedBoxesImagePath: Path
)

val blackListed = Set[String](
  "ai.djl.paddlepaddle:face_detection:0.0.1", // java.lang.IndexOutOfBoundsException: Incorrect number of elements in NDList.singletonOrThrow: Expected 1 and was 4
  "ai.djl.zoo:ssd:0.0.2",                     // java.lang.ArrayIndexOutOfBoundsException: Index 1 out of bounds for length 1
)

def testModel(modelArtifact: ModelArtifact, inputImageSources: List[URL]): List[ModelResult] = {
  println(s"${RED}TESTING MODEL $modelArtifact$RESET")
  val criteria =
    Criteria
      .builder()
      .setTypes(classOf[Image], classOf[DetectedObjects])
      .optApplication(Application.CV.OBJECT_DETECTION)
      .optGroupId(modelArtifact.groupId)
      .optArtifactId(modelArtifact.artifactId)
      .optFilters(modelArtifact.properties.asJava)
      .optProgress(new ProgressBar)
      .build()

  try {
    val model     = ModelZoo.loadModel(criteria)
    val predictor = model.newPredictor()

    inputImageSources.map { inputImageSource =>

      val inputImage                = ImageFactory.getInstance().fromUrl(inputImageSource)
      val started                   = System.currentTimeMillis()
      val detected: DetectedObjects = predictor.predict(inputImage)
      val duration                  = Duration.apply(System.currentTimeMillis() - started, TimeUnit.MILLISECONDS)
      val detectedObjects           = detected
        .items[DetectedObject]()
        .asScala
        .toList

      val outputImageFile = outputDir.resolve(s"${basename(inputImageSource.getFile)}-${modelArtifact.uuid}.png")
      saveBoundingBoxImage(inputImage, detected, outputImageFile)

      ModelResult(
        inputImageSource = inputImageSource,
        modelArtifact = modelArtifact,
        selectedModelPath = model.getModelPath,
        responseTime = duration,
        detectedObjects = detectedObjects,
        generatedBoundedBoxesImagePath = outputImageFile
      )
    }
  } catch {
    case err: ModelNotFoundException =>
      println(s"No matching model for $modelArtifact : ${err.getMessage}")
      Nil
  }
}

def showResults(results: Seq[ModelResult]): Unit = {
  results.groupBy(_.inputImageSource).foreach { (imageURL, resultsForImage) =>
    println(s"${BLUE}${BOLD}==========================================================================$RESET")
    println(s"${BLUE}${BOLD}RESULTS FOR $imageURL$RESET")
    resultsForImage.foreach { result =>
      import result.*
      println(s"${BLUE}${BOLD}--------------------------------------------------------------------------$RESET")
      println(s"${BLUE}${BOLD}MODEL ${modelArtifact.ident}$RESET")
      println(s"${BLUE}PATH $selectedModelPath$RESET")
      println(s"${GREEN}Number of detected object : ${detectedObjects.size} in $responseTime$RESET")
      println(s"${GREEN}  look at $generatedBoundedBoxesImagePath$RESET")
      detectedObjects.sortBy(-_.getProbability).foreach { detectedObject =>
        println(f"    $YELLOW$BOLD${detectedObject.getClassName} ${RED} ${detectedObject.getProbability}%1.2f$RESET")
      }
    }
  }
}

val inputImageSources =
  1.to(16).toList.map(n => URI.create(f"https://data.code-examples.org/ai/images-samples/example-$n%03d.jpg").toURL)

val objectDetectionsArtifacts =
  ModelZoo
    .listModels()
    .asScala
    .get(Application.CV.OBJECT_DETECTION)
    .map(_.asScala)
    .getOrElse(Nil)
    .toList

val results =
  objectDetectionsArtifacts
    .map(ModelArtifact.apply)
    .filterNot(artifactKey => blackListed.contains(artifactKey.ident))
    .map(modelArtifact => Try(testModel(modelArtifact, inputImageSources)))
    .collect{case Success(results) => results} // TODO just ignoring failing models...
    .flatten

showResults(results)