/JavaTF2ModelRunner

a demo of using tensorflow to run a model on an image.

Primary LanguageJavaMIT LicenseMIT

JavaTF2ModelRunner

App

a demo of using tensorflow to run a model on an image.

To run this, compile with maven as normal.

mvn package

Then execute the main class.

mvn exec:java -D"exec.mainClass"="org.orangepalantir.App" -Dexec.args="$MODEL $IMAGE"

FloatPredictor

Abstracts away the ImagePlus. Images are a float array with associated channels, slices, height and width.

PredictionServer and PredictionClient

A very simple socket server client implementation for sending and recieving images and predictions via a socket.

Troubleshooting

This error seems to happen when I don't have enough memory for the model.

Failed to get convolution algorithm. This is probably because cuDNN failed to initialize,

To address that I set the env variable TF_FORCE_GPU_ALLOW_GROWTH