Deep Java Library

Since Camel 3.3

Only producer is supported

The Deep Java Library component is used to infer deep learning models from message exchanges data. This component uses the Deep Java Library as the underlying library.

To use the DJL component, Maven users will need to add the following dependency to their pom.xml:

<dependency>
    <groupId>org.apache.camel</groupId>
    <artifactId>camel-djl</artifactId>
    <version>x.x.x</version>
    <!-- use the same version as your Camel core version -->
</dependency>

URI format

djl:application

Where application represents the application in the context of DJL, the common functional signature for a group of deep learning models.

Supported applications

Currently, the following applications are supported by the component.

Application Input types Output type

cv/image_classification

ai.djl.modality.cv.Image
byte[]
InputStream
File

ai.djl.modality.Classifications

cv/object_detection

ai.djl.modality.cv.Image
byte[]
InputStream
File

ai.djl.modality.cv.output.DetectedObjects

cv/semantic_segmentation

ai.djl.modality.cv.Image
byte[]
InputStream
File

ai.djl.modality.cv.output.CategoryMask

cv/instance_segmentation

ai.djl.modality.cv.Image
byte[]
InputStream
File

ai.djl.modality.cv.output.DetectedObjects

cv/pose_estimation

ai.djl.modality.cv.Image
byte[]
InputStream
File

ai.djl.modality.cv.output.Joints

cv/action_recognition

ai.djl.modality.cv.Image
byte[]
InputStream
File

ai.djl.modality.Classifications

cv/word_recognition

ai.djl.modality.cv.Image
byte[]
InputStream
File

String

cv/image_generation

int[]

ai.djl.modality.cv.Image[]

cv/image_enhancement

ai.djl.modality.cv.Image
byte[]
InputStream
File

ai.djl.modality.cv.Image

nlp/fill_mask

String

String[]

nlp/question_answer

ai.djl.modality.nlp.qa.QAInput
String[]

String

nlp/text_classification

String

ai.djl.modality.Classifications

nlp/sentiment_analysis

String

ai.djl.modality.Classifications

nlp/token_classification

String

ai.djl.modality.Classifications

nlp/text_generation

String

String

nlp/machine_translation

String

String

nlp/multiple_choice

String

String

nlp/text_embedding

String

ai.djl.ndarray.NDArray

audio

ai.djl.modality.audio.Audio
byte[]
InputStream
File

String

timeseries/forecasting

ai.djl.timeseries.TimeSeriesData

ai.djl.timeseries.Forecast

Configuring Options

Camel components are configured on two separate levels:

  • component level

  • endpoint level

Configuring Component Options

At the component level, you set general and shared configurations that are, then, inherited by the endpoints. It is the highest configuration level.

For example, a component may have security settings, credentials for authentication, urls for network connection and so forth.

Some components only have a few options, and others may have many. Because components typically have pre-configured defaults that are commonly used, then you may often only need to configure a few options on a component; or none at all.

You can configure components using:

  • the Component DSL.

  • in a configuration file (application.properties, *.yaml files, etc).

  • directly in the Java code.

Configuring Endpoint Options

You usually spend more time setting up endpoints because they have many options. These options help you customize what you want the endpoint to do. The options are also categorized into whether the endpoint is used as a consumer (from), as a producer (to), or both.

Configuring endpoints is most often done directly in the endpoint URI as path and query parameters. You can also use the Endpoint DSL and DataFormat DSL as a type safe way of configuring endpoints and data formats in Java.

A good practice when configuring options is to use Property Placeholders.

Property placeholders provide a few benefits:

  • They help prevent using hardcoded urls, port numbers, sensitive information, and other settings.

  • They allow externalizing the configuration from the code.

  • They help the code to become more flexible and reusable.

The following two sections list all the options, firstly for the component followed by the endpoint.

Component Options

The Deep Java Library component supports 2 options, which are listed below.

Name Description Default Type

lazyStartProducer (producer)

Whether the producer should be started lazy (on the first message). By starting lazy you can use this to allow CamelContext and routes to startup in situations where a producer may otherwise fail during starting and cause the route to fail being started. By deferring this startup to be lazy then the startup failure can be handled during routing messages via Camel’s routing error handlers. Beware that when the first message is processed then creating and starting the producer may take a little time and prolong the total processing time of the processing.

false

boolean

autowiredEnabled (advanced)

Whether autowiring is enabled. This is used for automatic autowiring options (the option must be marked as autowired) by looking up in the registry to find if there is a single instance of matching type, which then gets configured on the component. This can be used for automatic configuring JDBC data sources, JMS connection factories, AWS Clients, etc.

true

boolean

Endpoint Options

The Deep Java Library endpoint is configured using URI syntax:

djl:application

With the following path and query parameters:

Path Parameters (1 parameters)

Name Description Default Type

application (producer)

Required Application name.

String

Query Parameters (5 parameters)

Name Description Default Type

artifactId (producer)

Model Artifact.

String

model (producer)

Model.

String

showProgress (producer)

Show progress while loading zoo models. This parameter takes effect only with zoo models.

false

boolean

translator (producer)

Translator.

String

lazyStartProducer (producer (advanced))

Whether the producer should be started lazy (on the first message). By starting lazy you can use this to allow CamelContext and routes to startup in situations where a producer may otherwise fail during starting and cause the route to fail being started. By deferring this startup to be lazy then the startup failure can be handled during routing messages via Camel’s routing error handlers. Beware that when the first message is processed then creating and starting the producer may take a little time and prolong the total processing time of the processing.

false

boolean

Message Headers

The Deep Java Library component supports 1 message header(s), which is/are listed below:

Name Description Default Type

CamelDjlInput (producer)

Constant: INPUT

The input data used for prediction.

Model Zoo

The following tables contain supported models in the model zoos per application.

Those applications without a table mean that there are no pre-trained models found for them from the basic, PyTorch, TensorFlow or MXNet DJL model zoos. You may still find more models for an application from other model zoos such as Hugging Face, ONNX, etc.

CV - Image Classification

Application: cv/image_classification

Model family Artifact ID Options

MLP

ai.djl.zoo:mlp:0.0.3

{dataset=mnist}

MLP

ai.djl.mxnet:mlp:0.0.1

{dataset=mnist}

ResNet

ai.djl.zoo:resnet:0.0.2

{layers=50, flavor=v1, dataset=cifar10}

ResNet

ai.djl.pytorch:resnet:0.0.1

{layers=50, dataset=imagenet}
{layers=18, dataset=imagenet}
{layers=101, dataset=imagenet}

ResNet

ai.djl.tensorflow:resnet:0.0.1

{flavor=v1, layers=50, dataset=imagenet}

ResNet

ai.djl.mxnet:resnet:0.0.1

{layers=18, flavor=v1, dataset=imagenet}
{layers=50, flavor=v2, dataset=imagenet}
{layers=101, dataset=imagenet}
{layers=152, flavor=v1d, dataset=imagenet}
{layers=50, flavor=v1, dataset=cifar10}

ResNet-18

ai.djl.pytorch:resnet18_embedding:0.0.1

{}

SENet

ai.djl.mxnet:senet:0.0.1

{layers=154, dataset=imagenet}

SE-ResNeXt

ai.djl.mxnet:se_resnext:0.0.1

{layers=101, flavor=32x4d, dataset=imagenet}
{layers=101, flavor=64x4d, dataset=imagenet}

ResNeSt

ai.djl.mxnet:resnest:0.0.1

{layers=14, dataset=imagenet}
{layers=26, dataset=imagenet}
{layers=50, dataset=imagenet}
{layers=101, dataset=imagenet}
{layers=200, dataset=imagenet}
{layers=269, dataset=imagenet}

SqueezeNet

ai.djl.mxnet:squeezenet:0.0.1

{flavor=1.0, dataset=imagenet}

MobileNet

ai.djl.tensorflow:mobilenet:0.0.1

{flavor=v2, dataset=imagenet}

MobileNet

ai.djl.mxnet:mobilenet:0.0.1

{flavor=v1, multiplier=0.25, dataset=imagenet}
{flavor=v1, multiplier=0.5, dataset=imagenet}
{flavor=v1, multiplier=0.75, dataset=imagenet}
{flavor=v1, multiplier=1.0, dataset=imagenet}
{flavor=v2, multiplier=0.25, dataset=imagenet}
{flavor=v2, multiplier=0.5, dataset=imagenet}
{flavor=v2, multiplier=0.75, dataset=imagenet}
{flavor=v2, multiplier=1.0, dataset=imagenet}
{flavor=v3_small, multiplier=1.0, dataset=imagenet}
{flavor=v3_large, multiplier=1.0, dataset=imagenet}

GoogLeNet

ai.djl.mxnet:googlenet:0.0.1

{dataset=imagenet}

Darknet

ai.djl.mxnet:darknet:0.0.1

{layers=53, flavor=v3, dataset=imagenet}

Inception v3

ai.djl.mxnet:inceptionv3:0.0.1

{dataset=imagenet}

AlexNet

ai.djl.mxnet:alexnet:0.0.1

{dataset=imagenet}

VGGNet

ai.djl.mxnet:vgg:0.0.1

{layers=11, dataset=imagenet}
{layers=13, dataset=imagenet}
{layers=16, dataset=imagenet}
{layers=19, dataset=imagenet}
{flavor=batch_norm, layers=11, dataset=imagenet}
{flavor=batch_norm, layers=13, dataset=imagenet}
{flavor=batch_norm, layers=16, dataset=imagenet}
{flavor=batch_norm, layers=19, dataset=imagenet}

DenseNet

ai.djl.mxnet:densenet:0.0.1

{layers=121, dataset=imagenet}
{layers=161, dataset=imagenet}
{layers=169, dataset=imagenet}
{layers=201, dataset=imagenet}

Xception

ai.djl.mxnet:xception:0.0.1

{flavor=65, dataset=imagenet}

CV - Object Detection

Application: cv/object_detection

Model family Artifact ID Options

SSD

ai.djl.zoo:ssd:0.0.2

{flavor=tiny, dataset=pikachu}

SSD

ai.djl.pytorch:ssd:0.0.1

{size=300, backbone=resnet50, dataset=coco}

SSD

ai.djl.tensorflow:ssd:0.0.1

{backbone=mobilenet_v2, dataset=openimages_v4}

SSD

ai.djl.mxnet:ssd:0.0.1

{size=512, backbone=resnet50, flavor=v1, dataset=voc}
{size=512, backbone=vgg16, flavor=atrous, dataset=coco}
{size=512, backbone=mobilenet1.0, dataset=voc}
{size=300, backbone=vgg16, flavor=atrous, dataset=voc}

YOLO

ai.djl.mxnet:yolo:0.0.1

{dataset=voc, version=3, backbone=darknet53, imageSize=320}
{dataset=voc, version=3, backbone=darknet53, imageSize=416}
{dataset=voc, version=3, backbone=mobilenet1.0, imageSize=320}
{dataset=voc, version=3, backbone=mobilenet1.0, imageSize=416}
{dataset=coco, version=3, backbone=darknet53, imageSize=320}
{dataset=coco, version=3, backbone=darknet53, imageSize=416}
{dataset=coco, version=3, backbone=darknet53, imageSize=608}
{dataset=coco, version=3, backbone=mobilenet1.0, imageSize=320}
{dataset=coco, version=3, backbone=mobilenet1.0, imageSize=416}
{dataset=coco, version=3, backbone=mobilenet1.0, imageSize=608}

YOLOv5

ai.djl.pytorch:yolo5s:0.0.1

{}

YOLOv8

ai.djl.pytorch:yolov8n:0.0.1

{}

CV - Semantic Segmentation

Application: cv/semantic_segmentation

Model family Artifact ID Options

DeepLabV3

ai.djl.pytorch:deeplabv3:0.0.1

{backbone=resnet50, flavor=v1b, dataset=coco}

CV - Instance Segmentation

Application: cv/instance_segmentation

Model family Artifact ID Options

Mask R-CNN

ai.djl.mxnet:mask_rcnn:0.0.1

{backbone=resnet18, flavor=v1b, dataset=coco}
{backbone=resnet101, flavor=v1d, dataset=coco}

CV - Pose Estimation

Application: cv/pose_estimation

Model family Artifact ID Options

Simple Pose

ai.djl.mxnet:simple_pose:0.0.1

{backbone=resnet18, flavor=v1b, dataset=imagenet}
{backbone=resnet50, flavor=v1b, dataset=imagenet}
{backbone=resnet101, flavor=v1d, dataset=imagenet}
{backbone=resnet152, flavor=v1b, dataset=imagenet}
{backbone=resnet152, flavor=v1d, dataset=imagenet}

CV - Action Recognition

Application: cv/action_recognition

Model family Artifact ID Options

Action Recognition

ai.djl.mxnet:action_recognition:0.0.1

{backbone=vgg16, dataset=ucf101}
{backbone=inceptionv3, dataset=ucf101}

CV - Image Generation

Application: cv/image_generation

Model family Artifact ID Options

CycleGAN

ai.djl.pytorch:cyclegan:0.0.1

{artist=cezanne}
{artist=monet}
{artist=ukiyoe}
{artist=vangogh}

BigGAN

ai.djl.pytorch:biggan-deep:0.0.1

{layers=12, size=128, dataset=imagenet}
{layers=24, size=256, dataset=imagenet}
{layers=12, size=512, dataset=imagenet}

NLP - Question Answer

Application: nlp/question_answer

Model family Artifact ID Options

BertQA

ai.djl.pytorch:bertqa:0.0.1

{modelType=distilbert, size=base, cased=false, dataset=SQuAD}
{modelType=distilbert, size=base, cased=true, dataset=SQuAD}
{backbone=bert, cased=false, dataset=SQuAD}
{backbone=bert, cased=true, dataset=SQuAD}
{backbone=distilbert, cased=true, dataset=SQuAD}

BertQA

ai.djl.mxnet:bertqa:0.0.1

{backbone=bert, dataset=book_corpus_wiki_en_uncased}

NLP - Sentiment Analysis

Application: nlp/sentiment_analysis

Model family Artifact ID Options

DistilBERT

ai.djl.pytorch:distilbert:0.0.1

{backbone=distilbert, dataset=sst}

NLP - Word Embedding

Application: nlp/word_embedding

Model family Artifact ID Options

GloVe

ai.djl.mxnet:glove:0.0.2

{dimensions=50}

Time Series - Forecasting

Application: timeseries/forecasting

Model family Artifact ID Options

DeepAR

ai.djl.pytorch:deepar:0.0.1

{dataset=m5forecast}

DeepAR

ai.djl.mxnet:deepar:0.0.1

{dataset=airpassengers}
{dataset=m5forecast}

DJL Engine implementation

Because DJL is deep learning framework-agnostic, you don’t have to make a choice between frameworks when creating your projects. You can switch frameworks at any point. To ensure the best performance, DJL also provides automatic CPU/GPU choice based on hardware configuration.

PyTorch engine

You can pull the PyTorch engine from the central Maven repository by including the following dependency:

<dependency>
    <groupId>ai.djl.pytorch</groupId>
    <artifactId>pytorch-engine</artifactId>
    <version>x.x.x</version>
    <scope>runtime</scope>
</dependency>

By default, DJL will download the PyTorch native libraries into cache folder the first time you run DJL. It will automatically determine the appropriate jars for your system based on the platform and GPU support.

More information about PyTorch engine installation

TensorFlow engine

You can pull the TensorFlow engine from the central Maven repository by including the following dependency:

<dependency>
    <groupId>ai.djl.tensorflow</groupId>
    <artifactId>tensorflow-engine</artifactId>
    <version>x.x.x</version>
    <scope>runtime</scope>
</dependency>

By default, DJL will download the TensorFlow native libraries into cache folder the first time you run DJL. It will automatically determine the appropriate jars for your system based on the platform and GPU support.

More information about TensorFlow engine installation

MXNet engine

You can pull the MXNet engine from the central Maven repository by including the following dependency:

<dependency>
    <groupId>ai.djl.mxnet</groupId>
    <artifactId>mxnet-engine</artifactId>
    <version>x.x.x</version>
    <scope>runtime</scope>
</dependency>

By default, DJL will download the Apache MXNet native libraries into cache folder the first time you run DJL. It will automatically determine the appropriate jars for your system based on the platform and GPU support.

More information about MXNet engine installation

Examples

MNIST image classification from file

from("file:/data/mnist/0/10.png")
    .convertBodyTo(byte[].class)
    .to("djl:cv/image_classification?artifactId=ai.djl.mxnet:mlp:0.0.1");

Object detection

from("file:/data/mnist/0/10.png")
    .convertBodyTo(byte[].class)
    .to("djl:cv/image_classification?artifactId=ai.djl.mxnet:mlp:0.0.1");

Custom deep learning model

// create a deep learning model
Model model = Model.newInstance();
model.setBlock(new Mlp(28 * 28, 10, new int[]{128, 64}));
model.load(Paths.get(MODEL_DIR), MODEL_NAME);

// create translator for pre-processing and postprocessing
ImageClassificationTranslator.Builder builder = ImageClassificationTranslator.builder();
builder.setSynsetArtifactName("synset.txt");
builder.setPipeline(new Pipeline(new ToTensor()));
builder.optApplySoftmax(true);
ImageClassificationTranslator translator = new ImageClassificationTranslator(builder);

// Bind model and translator beans
context.getRegistry().bind("MyModel", model);
context.getRegistry().bind("MyTranslator", translator);

from("file:/data/mnist/0/10.png")
    .convertBodyTo(byte[].class)
    .to("djl:cv/image_classification?model=MyModel&translator=MyTranslator");

Spring Boot Auto-Configuration

When using djl with Spring Boot make sure to use the following Maven dependency to have support for auto configuration:

<dependency>
  <groupId>org.apache.camel.springboot</groupId>
  <artifactId>camel-djl-starter</artifactId>
  <version>x.x.x</version>
  <!-- use the same version as your Camel core version -->
</dependency>

The component supports 3 options, which are listed below.

Name Description Default Type

camel.component.djl.autowired-enabled

Whether autowiring is enabled. This is used for automatic autowiring options (the option must be marked as autowired) by looking up in the registry to find if there is a single instance of matching type, which then gets configured on the component. This can be used for automatic configuring JDBC data sources, JMS connection factories, AWS Clients, etc.

true

Boolean

camel.component.djl.enabled

Whether to enable auto configuration of the djl component. This is enabled by default.

Boolean

camel.component.djl.lazy-start-producer

Whether the producer should be started lazy (on the first message). By starting lazy you can use this to allow CamelContext and routes to startup in situations where a producer may otherwise fail during starting and cause the route to fail being started. By deferring this startup to be lazy then the startup failure can be handled during routing messages via Camel’s routing error handlers. Beware that when the first message is processed then creating and starting the producer may take a little time and prolong the total processing time of the processing.

false

Boolean
Back to top