Zero to image recognition in 60 seconds with Tensorflow and Spring Boot

TL;DR: for the impatient ones wanting to have a web service for image recognition without any Tensorflow prerequisites: run ./gradlew fetchInceptionFrozenModel bootrun, navigate to http://localhost:8080 and upload an image. The backend will categorize the image and output the result along with the probability.

Screenshot (non clickable)

Why

Tensorflow is hard enough to wrap one's head around. It has several parts that deal with preparing data, defining and training a model and finally, outputting a model that can then be used to categorize (infer) other data. There's math involved, new vocabulary to learn and on top, a toolchain which revolves around Python. This project only addresses serving a Tensorflow pre-trained image categorization model, otherwise called the Inception model.

Prerequisites

• JDK 17-21

Run

./gradlew fetchInceptionFrozenModel bootrun

Or build & run executable jar

./gradlew bootjar
build/libs/inception-serving-sb.jar

Why it doesn't this work on my Apple Silicon machine?

Tensorflow jars are not distributed for the Apple Silicon hardware. More info here

Then navigate to http://localhost:8080 and upload an image. The backend will categorize the image and output the result along with the probability.

How

Head to the blog post for the ful monty.

Kotlin

Thanks to @mcjojos, a port to Kotlin is now available in a separate branch. Woot!