4992350528

Build Type

github

Committed by GitHub

Commit Message

Merge 424c7ff18 into ef7906c5e

Pull Request Pull Request #13797: SPARKNLP-835: ProtectedParam and ProtectedFeature

Run Details

24 of 24 new or added lines in 6 files covered. (100.0%)

8643 of 13129 relevant lines covered (65.83%)

0.66 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

0.0

/src/main/scala/com/johnsnowlabs/nlp/annotators/cv/ViTForImageClassification.scala

/*
 * Copyright 2017-2022 John Snow Labs
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *    http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

package com.johnsnowlabs.nlp.annotators.cv

import com.johnsnowlabs.ml.ai.ViTClassifier
import com.johnsnowlabs.ml.tensorflow.{
  ReadTensorflowModel,
  TensorflowWrapper,
  WriteTensorflowModel
}
import com.johnsnowlabs.ml.util.LoadExternalModel.{
  loadJsonStringAsset,
  modelSanityCheck,
  notSupportedEngineError
}
import com.johnsnowlabs.ml.util.ModelEngine
import com.johnsnowlabs.nlp.AnnotatorType.{CATEGORY, IMAGE}
import com.johnsnowlabs.nlp._
import com.johnsnowlabs.nlp.annotators.cv.feature_extractor.Preprocessor
import com.johnsnowlabs.nlp.serialization.MapFeature
import org.apache.spark.broadcast.Broadcast
import org.apache.spark.ml.param.IntArrayParam
import org.apache.spark.ml.util.Identifiable
import org.apache.spark.sql.SparkSession
import org.json4s._
import org.json4s.jackson.JsonMethods._

import java.io.File

/** Vision Transformer (ViT) for image classification.
  *
  * ViT is a transformer based alternative to the convolutional neural networks usually used for
  * image recognition tasks.
  *
  * Pretrained models can be loaded with `pretrained` of the companion object:
  * {{{
  * val imageClassifier = ViTForImageClassification.pretrained()
  *   .setInputCols("image_assembler")
  *   .setOutputCol("class")
  * }}}
  * The default model is `"image_classifier_vit_base_patch16_224"`, if no name is provided.
  *
  * For available pretrained models please see the
  * [[https://sparknlp.org/models?task=Image+Classification Models Hub]].
  *
  * Models from the HuggingFace 🤗 Transformers library are also compatible with Spark NLP 🚀. To
  * see which models are compatible and how to import them see
  * [[https://github.com/JohnSnowLabs/spark-nlp/discussions/5669]] and to see more extended
  * examples, see
  * [[https://github.com/JohnSnowLabs/spark-nlp/blob/master/src/test/scala/com/johnsnowlabs/nlp/annotators/cv/ViTImageClassificationTestSpec.scala ViTImageClassificationTestSpec]].
  *
  * '''References:'''
  *
  * [[https://arxiv.org/abs/2010.11929 An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale]]
  *
  * '''Paper Abstract:'''
  *
  * ''While the Transformer architecture has become the de-facto standard for natural language
  * processing tasks, its applications to computer vision remain limited. In vision, attention is
  * either applied in conjunction with convolutional networks, or used to replace certain
  * components of convolutional networks while keeping their overall structure in place. We show
  * that this reliance on CNNs is not necessary and a pure transformer applied directly to
  * sequences of image patches can perform very well on image classification tasks. When
  * pre-trained on large amounts of data and transferred to multiple mid-sized or small image
  * recognition benchmarks (ImageNet, CIFAR-100, VTAB, etc.), Vision Transformer (ViT) attains
  * excellent results compared to state-of-the-art convolutional networks while requiring
  * substantially fewer computational resources to train.''
  *
  * ==Example==
  * {{{
  * import com.johnsnowlabs.nlp.annotator._
  * import com.johnsnowlabs.nlp.ImageAssembler
  * import org.apache.spark.ml.Pipeline
  *
  * val imageDF: DataFrame = spark.read
  *   .format("image")
  *   .option("dropInvalid", value = true)
  *   .load("src/test/resources/image/")
  *
  * val imageAssembler = new ImageAssembler()
  *   .setInputCol("image")
  *   .setOutputCol("image_assembler")
  *
  * val imageClassifier = ViTForImageClassification
  *   .pretrained()
  *   .setInputCols("image_assembler")
  *   .setOutputCol("class")
  *
  * val pipeline = new Pipeline().setStages(Array(imageAssembler, imageClassifier))
  * val pipelineDF = pipeline.fit(imageDF).transform(imageDF)
  *
  * pipelineDF
  *   .selectExpr("reverse(split(image.origin, '/'))[0] as image_name", "class.result")
  *   .show(truncate = false)
  * +-----------------+----------------------------------------------------------+
  * |image_name       |result                                                    |
  * +-----------------+----------------------------------------------------------+
  * |palace.JPEG      |[palace]                                                  |
  * |egyptian_cat.jpeg|[Egyptian cat]                                            |
  * |hippopotamus.JPEG|[hippopotamus, hippo, river horse, Hippopotamus amphibius]|
  * |hen.JPEG         |[hen]                                                     |
  * |ostrich.JPEG     |[ostrich, Struthio camelus]                               |
  * |junco.JPEG       |[junco, snowbird]                                         |
  * |bluetick.jpg     |[bluetick]                                                |
  * |chihuahua.jpg    |[Chihuahua]                                               |
  * |tractor.JPEG     |[tractor]                                                 |
  * |ox.JPEG          |[ox]                                                      |
  * +-----------------+----------------------------------------------------------+
  * }}}
  *
  * @param uid
  *   required uid for storing annotator to disk
  * @groupname anno Annotator types
  * @groupdesc anno
  *   Required input and expected output annotator types
  * @groupname Ungrouped Members
  * @groupname param Parameters
  * @groupname setParam Parameter setters
  * @groupname getParam Parameter getters
  * @groupname Ungrouped Members
  * @groupprio param  1
  * @groupprio anno  2
  * @groupprio Ungrouped 3
  * @groupprio setParam  4
  * @groupprio getParam  5
  * @groupdesc param
  *   A list of (hyper-)parameter keys this annotator can take. Users can set and get the
  *   parameter values through setters and getters, respectively.
  */
class ViTForImageClassification(override val uid: String)
    extends AnnotatorModel[ViTForImageClassification]
    with HasBatchedAnnotateImage[ViTForImageClassification]
    with HasImageFeatureProperties
    with WriteTensorflowModel
    with HasEngine {

  /** Annotator reference id. Used to identify elements in metadata or to refer to this annotator
    * type
    */
  def this() = this(Identifiable.randomUID("ViTForImageClassification"))

  /** Output annotator type : CATEGORY
    *
    * @group anno
    */
  override val outputAnnotatorType: AnnotatorType = CATEGORY

  /** Input annotator type : IMAGE
    *
    * @group anno
    */
  override val inputAnnotatorTypes: Array[AnnotatorType] = Array(IMAGE)

  /** ConfigProto from tensorflow, serialized into byte array. Get with
    * config_proto.SerializeToString()
    *
    * @group param
    */
  val configProtoBytes = new IntArrayParam(
    this,
    "configProtoBytes",
    "ConfigProto from tensorflow, serialized into byte array. Get with config_proto.SerializeToString()")

  /** ConfigProto from tensorflow, serialized into byte array. Get with
    * config_proto.SerializeToString()
    *
    * @group setParam
    */
  def setConfigProtoBytes(bytes: Array[Int]): ViTForImageClassification.this.type =
    set(this.configProtoBytes, bytes)

  /** ConfigProto from tensorflow, serialized into byte array. Get with
    * config_proto.SerializeToString()
    *
    * @group getParam
    */
  def getConfigProtoBytes: Option[Array[Byte]] =
    get(this.configProtoBytes).map(_.map(_.toByte))

  /** Labels used to decode predicted IDs back to string tags
    *
    * @group param
    */
  val labels: MapFeature[String, BigInt] = new MapFeature(this, "labels").setProtected()

  /** @group setParam */
  def setLabels(value: Map[String, BigInt]): this.type = set(labels, value)

  /** Returns labels used to train this model */
  def getClasses: Array[String] = {
    $$(labels).keys.toArray
  }

  /** It contains TF model signatures for the laded saved model
    *
    * @group param
    */
  val signatures =
    new MapFeature[String, String](model = this, name = "signatures").setProtected()

  /** @group setParam */
  def setSignatures(value: Map[String, String]): this.type = {
    set(signatures, value)
    this
  }

  /** @group getParam */
  def getSignatures: Option[Map[String, String]] = get(this.signatures)

  private var _model: Option[Broadcast[ViTClassifier]] = None

  /** @group getParam */
  def getModelIfNotSet: ViTClassifier = _model.get.value

  /** @group setParam */
  def setModelIfNotSet(
      spark: SparkSession,
      tensorflow: TensorflowWrapper,
      preprocessor: Preprocessor): this.type = {
    if (_model.isEmpty) {

      _model = Some(
        spark.sparkContext.broadcast(
          new ViTClassifier(
            tensorflow,
            configProtoBytes = getConfigProtoBytes,
            tags = $$(labels),
            preprocessor = preprocessor,
            signatures = getSignatures)))
    }
    this
  }

  setDefault(batchSize -> 2)

  /** Takes a document and annotations and produces new annotations of this annotator's annotation
    * type
    *
    * @param batchedAnnotations
    *   Annotations that correspond to inputAnnotationCols generated by previous annotators if any
    * @return
    *   any number of annotations processed for every input annotation. Not necessary one to one
    *   relationship
    */
  override def batchAnnotate(
      batchedAnnotations: Seq[Array[AnnotationImage]]): Seq[Seq[Annotation]] = {

    // Zip annotations to the row it belongs to
    val imagesWithRow = batchedAnnotations.zipWithIndex
      .flatMap { case (annotations, i) => annotations.map(x => (x, i)) }

    val noneEmptyImages = imagesWithRow.map(_._1).filter(_.result.nonEmpty).toArray

    val allAnnotations =
      if (noneEmptyImages.nonEmpty) {
        getModelIfNotSet.predict(
          images = noneEmptyImages,
          batchSize = $(batchSize),
          preprocessor = Preprocessor(
            do_normalize = getDoNormalize,
            do_resize = getDoResize,
            feature_extractor_type = getFeatureExtractorType,
            image_mean = getImageMean,
            image_std = getImageStd,
            resample = getResample,
            size = getSize))
      } else {
        Seq.empty[Annotation]
      }

    // Group resulting annotations by rows. If there are not sentences in a given row, return empty sequence
    batchedAnnotations.indices.map(rowIndex => {
      val rowAnnotations = allAnnotations
        // zip each annotation with its corresponding row index
        .zip(imagesWithRow)
        // select the sentences belonging to the current row
        .filter(_._2._2 == rowIndex)
        // leave the annotation only
        .map(_._1)

      if (rowAnnotations.nonEmpty)
        rowAnnotations
      else
        Seq.empty[Annotation]
    })

  }

  override def onWrite(path: String, spark: SparkSession): Unit = {
    super.onWrite(path, spark)
    writeTensorflowModelV2(
      path,
      spark,
      getModelIfNotSet.tensorflowWrapper,
      "_image_classification",
      ViTForImageClassification.tfFile,
      configProtoBytes = getConfigProtoBytes)
  }

}

trait ReadablePretrainedViTForImageModel
    extends ParamsAndFeaturesReadable[ViTForImageClassification]
    with HasPretrained[ViTForImageClassification] {
  override val defaultModelName: Some[String] = Some("image_classifier_vit_base_patch16_224")

  /** Java compliant-overrides */
  override def pretrained(): ViTForImageClassification = super.pretrained()

  override def pretrained(name: String): ViTForImageClassification = super.pretrained(name)

  override def pretrained(name: String, lang: String): ViTForImageClassification =
    super.pretrained(name, lang)

  override def pretrained(
      name: String,
      lang: String,
      remoteLoc: String): ViTForImageClassification = super.pretrained(name, lang, remoteLoc)
}

trait ReadViTForImageDLModel extends ReadTensorflowModel {
  this: ParamsAndFeaturesReadable[ViTForImageClassification] =>

  override val tfFile: String = "image_classification_tensorflow"

  def readModel(instance: ViTForImageClassification, path: String, spark: SparkSession): Unit = {

    val tf = readTensorflowModel(path, spark, "_image_classification_tf", initAllTables = false)

    val preprocessor = Preprocessor(
      do_normalize = true,
      do_resize = true,
      "ViTFeatureExtractor",
      instance.getImageMean,
      instance.getImageStd,
      instance.getResample,
      instance.getSize)

    instance.setModelIfNotSet(spark, tf, preprocessor)
  }

  addReader(readModel)

  def loadSavedModel(modelPath: String, spark: SparkSession): ViTForImageClassification = {

    val (localModelPath, detectedEngine) = modelSanityCheck(modelPath)

    // TODO: sometimes results in [String, BigInt] where BigInt is actually a string
    val labelJsonContent = loadJsonStringAsset(localModelPath, "labels.json")
    val labelJsonMap =
      parse(labelJsonContent, useBigIntForLong = true).values
        .asInstanceOf[Map[String, BigInt]]

    val preprocessorConfigJsonContent =
      loadJsonStringAsset(localModelPath, "preprocessor_config.json")
    val preprocessorConfig =
      Preprocessor.loadPreprocessorConfig(preprocessorConfigJsonContent)

    /*Universal parameters for all engines*/
    val annotatorModel = new ViTForImageClassification()
      .setLabels(labelJsonMap)
      .setDoNormalize(preprocessorConfig.do_normalize)
      .setDoResize(preprocessorConfig.do_resize)
      .setFeatureExtractorType(preprocessorConfig.feature_extractor_type)
      .setImageMean(preprocessorConfig.image_mean)
      .setImageStd(preprocessorConfig.image_std)
      .setResample(preprocessorConfig.resample)
      .setSize(preprocessorConfig.size)

    annotatorModel.set(annotatorModel.engine, detectedEngine)

    detectedEngine match {
      case ModelEngine.tensorflow =>
        val (wrapper, signatures) =
          TensorflowWrapper.read(localModelPath, zipped = false, useBundle = true)

        val _signatures = signatures match {
          case Some(s) => s
          case None => throw new Exception("Cannot load signature definitions from model!")
        }

        /** the order of setSignatures is important if we use getSignatures inside
          * setModelIfNotSet
          */
        annotatorModel
          .setSignatures(_signatures)
          .setModelIfNotSet(spark, wrapper, preprocessorConfig)

      case _ =>
        throw new Exception(notSupportedEngineError)
    }

    annotatorModel
  }
}

/** This is the companion object of [[ViTForImageClassification]]. Please refer to that class for
  * the documentation.
  */
object ViTForImageClassification
    extends ReadablePretrainedViTForImageModel
    with ReadViTForImageDLModel

1	/*
2	* Copyright 2017-2022 John Snow Labs
3	*
4	* Licensed under the Apache License, Version 2.0 (the "License");
5	* you may not use this file except in compliance with the License.
6	* You may obtain a copy of the License at
7	*
8	* http://www.apache.org/licenses/LICENSE-2.0
9	*
10	* Unless required by applicable law or agreed to in writing, software
11	* distributed under the License is distributed on an "AS IS" BASIS,
12	* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13	* See the License for the specific language governing permissions and
14	* limitations under the License.
15	*/
16
17	package com.johnsnowlabs.nlp.annotators.cv
18
19	import com.johnsnowlabs.ml.ai.ViTClassifier
20	import com.johnsnowlabs.ml.tensorflow.{
21	ReadTensorflowModel,
22	TensorflowWrapper,
23	WriteTensorflowModel
24	}
25	import com.johnsnowlabs.ml.util.LoadExternalModel.{
26	loadJsonStringAsset,
27	modelSanityCheck,
28	notSupportedEngineError
29	}
30	import com.johnsnowlabs.ml.util.ModelEngine
31	import com.johnsnowlabs.nlp.AnnotatorType.{CATEGORY, IMAGE}
32	import com.johnsnowlabs.nlp._
33	import com.johnsnowlabs.nlp.annotators.cv.feature_extractor.Preprocessor
34	import com.johnsnowlabs.nlp.serialization.MapFeature
35	import org.apache.spark.broadcast.Broadcast
36	import org.apache.spark.ml.param.IntArrayParam
37	import org.apache.spark.ml.util.Identifiable
38	import org.apache.spark.sql.SparkSession
39	import org.json4s._
40	import org.json4s.jackson.JsonMethods._
41
42	import java.io.File
43
44	/** Vision Transformer (ViT) for image classification.
45	*
46	* ViT is a transformer based alternative to the convolutional neural networks usually used for
47	* image recognition tasks.
48	*
49	* Pretrained models can be loaded with `pretrained` of the companion object:
50	* {{{
51	* val imageClassifier = ViTForImageClassification.pretrained()
52	* .setInputCols("image_assembler")
53	* .setOutputCol("class")
54	* }}}
55	* The default model is `"image_classifier_vit_base_patch16_224"`, if no name is provided.
56	*
57	* For available pretrained models please see the
58	* [[https://sparknlp.org/models?task=Image+Classification Models Hub]].
59	*
60	* Models from the HuggingFace 🤗 Transformers library are also compatible with Spark NLP 🚀. To
61	* see which models are compatible and how to import them see
62	* [[https://github.com/JohnSnowLabs/spark-nlp/discussions/5669]] and to see more extended
63	* examples, see
64	* [[https://github.com/JohnSnowLabs/spark-nlp/blob/master/src/test/scala/com/johnsnowlabs/nlp/annotators/cv/ViTImageClassificationTestSpec.scala ViTImageClassificationTestSpec]].
65	*
66	* '''References:'''
67	*
68	* [[https://arxiv.org/abs/2010.11929 An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale]]
69	*
70	* '''Paper Abstract:'''
71	*
72	* ''While the Transformer architecture has become the de-facto standard for natural language
73	* processing tasks, its applications to computer vision remain limited. In vision, attention is
74	* either applied in conjunction with convolutional networks, or used to replace certain
75	* components of convolutional networks while keeping their overall structure in place. We show
76	* that this reliance on CNNs is not necessary and a pure transformer applied directly to
77	* sequences of image patches can perform very well on image classification tasks. When
78	* pre-trained on large amounts of data and transferred to multiple mid-sized or small image
79	* recognition benchmarks (ImageNet, CIFAR-100, VTAB, etc.), Vision Transformer (ViT) attains
80	* excellent results compared to state-of-the-art convolutional networks while requiring
81	* substantially fewer computational resources to train.''
82	*
83	* ==Example==
84	* {{{
85	* import com.johnsnowlabs.nlp.annotator._
86	* import com.johnsnowlabs.nlp.ImageAssembler
87	* import org.apache.spark.ml.Pipeline
88	*
89	* val imageDF: DataFrame = spark.read
90	* .format("image")
91	* .option("dropInvalid", value = true)
92	* .load("src/test/resources/image/")
93	*
94	* val imageAssembler = new ImageAssembler()
95	* .setInputCol("image")
96	* .setOutputCol("image_assembler")
97	*
98	* val imageClassifier = ViTForImageClassification
99	* .pretrained()
100	* .setInputCols("image_assembler")
101	* .setOutputCol("class")
102	*
103	* val pipeline = new Pipeline().setStages(Array(imageAssembler, imageClassifier))
104	* val pipelineDF = pipeline.fit(imageDF).transform(imageDF)
105	*
106	* pipelineDF
107	* .selectExpr("reverse(split(image.origin, '/'))[0] as image_name", "class.result")
108	* .show(truncate = false)
109	* +-----------------+----------------------------------------------------------+
110	* \|image_name \|result \|
111	* +-----------------+----------------------------------------------------------+
112	* \|palace.JPEG \|[palace] \|
113	* \|egyptian_cat.jpeg\|[Egyptian cat] \|
114	* \|hippopotamus.JPEG\|[hippopotamus, hippo, river horse, Hippopotamus amphibius]\|
115	* \|hen.JPEG \|[hen] \|
116	* \|ostrich.JPEG \|[ostrich, Struthio camelus] \|
117	* \|junco.JPEG \|[junco, snowbird] \|
118	* \|bluetick.jpg \|[bluetick] \|
119	* \|chihuahua.jpg \|[Chihuahua] \|
120	* \|tractor.JPEG \|[tractor] \|
121	* \|ox.JPEG \|[ox] \|
122	* +-----------------+----------------------------------------------------------+
123	* }}}
124	*
125	* @param uid
126	* required uid for storing annotator to disk
127	* @groupname anno Annotator types
128	* @groupdesc anno
129	* Required input and expected output annotator types
130	* @groupname Ungrouped Members
131	* @groupname param Parameters
132	* @groupname setParam Parameter setters
133	* @groupname getParam Parameter getters
134	* @groupname Ungrouped Members
135	* @groupprio param 1
136	* @groupprio anno 2
137	* @groupprio Ungrouped 3
138	* @groupprio setParam 4
139	* @groupprio getParam 5
140	* @groupdesc param
141	* A list of (hyper-)parameter keys this annotator can take. Users can set and get the
142	* parameter values through setters and getters, respectively.
143	*/
144	class ViTForImageClassification(override val uid: String)
145	extends AnnotatorModel[ViTForImageClassification]
146	with HasBatchedAnnotateImage[ViTForImageClassification]
147	with HasImageFeatureProperties
148	with WriteTensorflowModel
149	with HasEngine {
150
151	/** Annotator reference id. Used to identify elements in metadata or to refer to this annotator
152	* type
153	*/
154	def this() = this(Identifiable.randomUID("ViTForImageClassification"))	×
155
156	/** Output annotator type : CATEGORY
157	*
158	* @group anno
159	*/
160	override val outputAnnotatorType: AnnotatorType = CATEGORY	×
161
162	/** Input annotator type : IMAGE
163	*
164	* @group anno
165	*/
166	override val inputAnnotatorTypes: Array[AnnotatorType] = Array(IMAGE)	×
167
168	/** ConfigProto from tensorflow, serialized into byte array. Get with
169	* config_proto.SerializeToString()
170	*
171	* @group param
172	*/
173	val configProtoBytes = new IntArrayParam(	×
174	this,
175	"configProtoBytes",	×
176	"ConfigProto from tensorflow, serialized into byte array. Get with config_proto.SerializeToString()")	×
177
178	/** ConfigProto from tensorflow, serialized into byte array. Get with
179	* config_proto.SerializeToString()
180	*
181	* @group setParam
182	*/
183	def setConfigProtoBytes(bytes: Array[Int]): ViTForImageClassification.this.type =
184	set(this.configProtoBytes, bytes)	×
185
186	/** ConfigProto from tensorflow, serialized into byte array. Get with
187	* config_proto.SerializeToString()
188	*
189	* @group getParam
190	*/
191	def getConfigProtoBytes: Option[Array[Byte]] =
192	get(this.configProtoBytes).map(_.map(_.toByte))	×
193
194	/** Labels used to decode predicted IDs back to string tags
195	*
196	* @group param
197	*/
198	val labels: MapFeature[String, BigInt] = new MapFeature(this, "labels").setProtected()	×
199
200	/** @group setParam */
201	def setLabels(value: Map[String, BigInt]): this.type = set(labels, value)	×
202
203	/** Returns labels used to train this model */
204	def getClasses: Array[String] = {
205	$$(labels).keys.toArray	×
206	}
207
208	/** It contains TF model signatures for the laded saved model
209	*
210	* @group param
211	*/
212	val signatures =
213	new MapFeature[String, String](model = this, name = "signatures").setProtected()	×
214
215	/** @group setParam */
216	def setSignatures(value: Map[String, String]): this.type = {
217	set(signatures, value)	×
218	this
219	}
220
221	/** @group getParam */
222	def getSignatures: Option[Map[String, String]] = get(this.signatures)	×
223
224	private var _model: Option[Broadcast[ViTClassifier]] = None	×
225
226	/** @group getParam */
227	def getModelIfNotSet: ViTClassifier = _model.get.value	×
228
229	/** @group setParam */
230	def setModelIfNotSet(
231	spark: SparkSession,
232	tensorflow: TensorflowWrapper,
233	preprocessor: Preprocessor): this.type = {
234	if (_model.isEmpty) {	×
235
236	_model = Some(	×
237	spark.sparkContext.broadcast(	×
238	new ViTClassifier(	×
239	tensorflow,
240	configProtoBytes = getConfigProtoBytes,	×
241	tags = $$(labels),	×
242	preprocessor = preprocessor,
243	signatures = getSignatures)))	×
244	}
245	this
246	}
247
248	setDefault(batchSize -> 2)	×
249
250	/** Takes a document and annotations and produces new annotations of this annotator's annotation
251	* type
252	*
253	* @param batchedAnnotations
254	* Annotations that correspond to inputAnnotationCols generated by previous annotators if any
255	* @return
256	* any number of annotations processed for every input annotation. Not necessary one to one
257	* relationship
258	*/
259	override def batchAnnotate(
260	batchedAnnotations: Seq[Array[AnnotationImage]]): Seq[Seq[Annotation]] = {
261
262	// Zip annotations to the row it belongs to
263	val imagesWithRow = batchedAnnotations.zipWithIndex	×
264	.flatMap { case (annotations, i) => annotations.map(x => (x, i)) }	×
265
266	val noneEmptyImages = imagesWithRow.map(_._1).filter(_.result.nonEmpty).toArray	×
267
268	val allAnnotations =
269	if (noneEmptyImages.nonEmpty) {	×
270	getModelIfNotSet.predict(	×
271	images = noneEmptyImages,
272	batchSize = $(batchSize),	×
273	preprocessor = Preprocessor(	×
274	do_normalize = getDoNormalize,	×
275	do_resize = getDoResize,	×
276	feature_extractor_type = getFeatureExtractorType,	×
277	image_mean = getImageMean,	×
278	image_std = getImageStd,	×
279	resample = getResample,	×
280	size = getSize))	×
281	} else {
282	Seq.empty[Annotation]	×
283	}
284
285	// Group resulting annotations by rows. If there are not sentences in a given row, return empty sequence
286	batchedAnnotations.indices.map(rowIndex => {	×
287	val rowAnnotations = allAnnotations
288	// zip each annotation with its corresponding row index
289	.zip(imagesWithRow)	×
290	// select the sentences belonging to the current row
291	.filter(_._2._2 == rowIndex)	×
292	// leave the annotation only
293	.map(_._1)	×
294
295	if (rowAnnotations.nonEmpty)	×
296	rowAnnotations	×
297	else
298	Seq.empty[Annotation]	×
299	})
300
301	}
302
303	override def onWrite(path: String, spark: SparkSession): Unit = {
304	super.onWrite(path, spark)	×
305	writeTensorflowModelV2(	×
306	path,
307	spark,
308	getModelIfNotSet.tensorflowWrapper,	×
309	"_image_classification",	×
310	ViTForImageClassification.tfFile,	×
311	configProtoBytes = getConfigProtoBytes)	×
312	}
313
314	}
315
316	trait ReadablePretrainedViTForImageModel
317	extends ParamsAndFeaturesReadable[ViTForImageClassification]
318	with HasPretrained[ViTForImageClassification] {
319	override val defaultModelName: Some[String] = Some("image_classifier_vit_base_patch16_224")	×
320
321	/** Java compliant-overrides */
322	override def pretrained(): ViTForImageClassification = super.pretrained()	×
323
324	override def pretrained(name: String): ViTForImageClassification = super.pretrained(name)	×
325
326	override def pretrained(name: String, lang: String): ViTForImageClassification =
327	super.pretrained(name, lang)	×
328
329	override def pretrained(
330	name: String,
331	lang: String,
332	remoteLoc: String): ViTForImageClassification = super.pretrained(name, lang, remoteLoc)	×
333	}
334
335	trait ReadViTForImageDLModel extends ReadTensorflowModel {
336	this: ParamsAndFeaturesReadable[ViTForImageClassification] =>
337
338	override val tfFile: String = "image_classification_tensorflow"	×
339
340	def readModel(instance: ViTForImageClassification, path: String, spark: SparkSession): Unit = {
341
342	val tf = readTensorflowModel(path, spark, "_image_classification_tf", initAllTables = false)	×
343
344	val preprocessor = Preprocessor(	×
345	do_normalize = true,	×
346	do_resize = true,	×
347	"ViTFeatureExtractor",	×
348	instance.getImageMean,	×
349	instance.getImageStd,	×
350	instance.getResample,	×
351	instance.getSize)	×
352
353	instance.setModelIfNotSet(spark, tf, preprocessor)	×
354	}
355
356	addReader(readModel)	×
357
358	def loadSavedModel(modelPath: String, spark: SparkSession): ViTForImageClassification = {
359
360	val (localModelPath, detectedEngine) = modelSanityCheck(modelPath)	×
361
362	// TODO: sometimes results in [String, BigInt] where BigInt is actually a string
363	val labelJsonContent = loadJsonStringAsset(localModelPath, "labels.json")	×
364	val labelJsonMap =
365	parse(labelJsonContent, useBigIntForLong = true).values	×
366	.asInstanceOf[Map[String, BigInt]]	×
367
368	val preprocessorConfigJsonContent =
369	loadJsonStringAsset(localModelPath, "preprocessor_config.json")	×
370	val preprocessorConfig =
371	Preprocessor.loadPreprocessorConfig(preprocessorConfigJsonContent)	×
372
373	/Universal parameters for all engines/
374	val annotatorModel = new ViTForImageClassification()
375	.setLabels(labelJsonMap)
376	.setDoNormalize(preprocessorConfig.do_normalize)	×
377	.setDoResize(preprocessorConfig.do_resize)	×
378	.setFeatureExtractorType(preprocessorConfig.feature_extractor_type)	×
379	.setImageMean(preprocessorConfig.image_mean)	×
380	.setImageStd(preprocessorConfig.image_std)	×
381	.setResample(preprocessorConfig.resample)	×
382	.setSize(preprocessorConfig.size)	×
383
384	annotatorModel.set(annotatorModel.engine, detectedEngine)	×
385
386	detectedEngine match {
387	case ModelEngine.tensorflow =>
388	val (wrapper, signatures) =	×
389	TensorflowWrapper.read(localModelPath, zipped = false, useBundle = true)
390
391	val _signatures = signatures match {
392	case Some(s) => s
393	case None => throw new Exception("Cannot load signature definitions from model!")	×
394	}
395
396	/** the order of setSignatures is important if we use getSignatures inside
397	* setModelIfNotSet
398	*/
399	annotatorModel
400	.setSignatures(_signatures)
401	.setModelIfNotSet(spark, wrapper, preprocessorConfig)	×
402
403	case _ =>
404	throw new Exception(notSupportedEngineError)	×
405	}
406
407	annotatorModel
408	}
409	}
410
411	/** This is the companion object of [[ViTForImageClassification]]. Please refer to that class for
412	* the documentation.
413	*/
414	object ViTForImageClassification
415	extends ReadablePretrainedViTForImageModel
416	with ReadViTForImageDLModel

JohnSnowLabs / spark-nlp / 4992350528

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous