Deep Java Library (DJL) Demo Project
The DJL Demo project is a showcase of practical applications and examples utilizing the Deep Java Library (DJL). It is a framework-agnostic Java API designed for deep learning tasks. This project repository serves as a comprehensive guide for developers, offering examples ranging from inference tasks to model training, Android applications, AWS service integration, and more.
Inference Examples
Run Python Pre/Post Processing
This example demonstrates how to execute Python code within the DJL framework, allowing seamless integration of Python scripts for preprocessing or postprocessing tasks in Java-based deep learning projects.
Malicious URL Detector
Explore a practical application that uses a Character Level CNN model to identify malicious URLs, providing insight into natural language processing capabilities within DJL.
Pneumonia Detection with TensorFlow
Utilizing a pretrained Keras model, this example illustrates how to detect Pneumonia from X-ray images using the TensorFlow engine, offering a blend of healthcare technology and machine learning.
Live Object Detection
Dive into computer vision by detecting live objects through a web camera feed, showcasing DJL’s real-time inference capabilities.
Online DoodleDraw Game
Experience the interactive, web-based DoodleDraw game powered by DJL, highlighting how deep learning models can be integrated into fun and engaging applications.
Deep Learning in Browser
A unique web application example that enables running DJL code directly in the browser, demonstrating the versatility and reach of deep learning applications outside traditional environments.
Training Examples
Train Footwear Classification
Learn how to train a model for classifying different types of footwear using DJL, a hands-on example illustrating model training and evaluation processes.
Visualizing Training with DJL
This example introduces a web UI that helps track and visualize training metrics like loss and accuracy, aiding in better understanding and monitoring of model performance.
Android Applications
Face Detection with Ease
A demonstration of building a deep learning Android app efficiently using DJL, focused on implementing face detection features.
Doodle Draw App
Explore an Android game built with a PyTorch model, illustrating creative and interactive mobile applications utilizing DJL.
Style Transfer
Enhance images by transforming them into the style of famous artists like Van Gogh or Monet, showcasing the artistic applications of deep learning in style transfer.
Semantic Segmentation
This Android application colors objects within an image, demonstrating object identification and semantic segmentation techniques.
Neural Machine Translation
Convert French text to English with an Android app, walking through the implementation of a neural machine translation system on a mobile platform.
Speech Recognition
Build a speech recognition app for Android, showcasing DJL’s capability to process and recognize spoken commands.
Object Detection with ONNX Model
A guide to developing an object detection app using an ONNX model, adding another layer of model deployment options for Android developers.
AWS Services Integration
AWS Kinesis Video Streams
This example reads outputs from a KVS Stream, integrating video processing capabilities with DJL’s deep learning features.
Serverless Model Serving with AWS Lambda
Implement a serverless approach to serving models using AWS Lambda, demonstrating scalability and flexibility in deploying deep learning models.
Model Serving on AWS Elastic Beanstalk
Learn how to develop a microservice for deploying models on AWS Elastic Beanstalk, integrating cloud services with DJL applications.
Low-Cost Inference with AWS Inferentia
Run high-performance, cost-effective inferences using AWS Inferentia, providing economic and efficient deployment strategies for deep learning models.
Big Data Integration
Spark Image Classification
A demonstration of image classification tasks using Spark, showing how DJL can be paired with big data frameworks for enhanced data processing.
Apache Beam CTR Prediction
Utilize Apache Beam to predict click-through rates for online advertisements, offering insights into stream processing and machine learning analytics.
Apache Flink Sentiment Analysis
Run sentiment analysis with Apache Flink, exploring real-time data processing and natural language understanding with DJL.
DJL Component in Apache Camel
Illustrates a simple HTTP service to classify images using the Zoo Model, showcasing integration points between DJL and Apache Camel for enterprise applications.
Other Demos
Multiple Engines in Single JVM
Explore how various deep learning frameworks can be executed within a single Java Virtual Machine, emphasizing DJL’s capability to harmonize across different platforms.
Speed Up with GraalVM
This example demonstrates how to speed up deep learning applications by compiling DJL apps into native executables using GraalVM, enhancing application performance.
Deploying DJL Models on Quarkus
Discusses serving deep learning models via Quarkus, highlighting modern approaches to deploying AI models in server-less environments.
Overall, the DJL Demo project offers an extensive collection of examples and applications that lower the barrier for Java developers looking to delve into the world of deep learning, providing accessible, actionable insights into the integration and deployment of AI technologies across various platforms and use cases.