tfrecord -> Rerun plugin

This is an example data-loader plugin that lets you view a TFRecord of Events (i.e., Tensorboard log files) in the Rerun Viewer. It uses the external data loader mechanism to add this capability to the viewer without modifying the viewer itself.

rerun_opens_tfevents.mp4

External data loaders are executables that are available to the Rerun Viewer via the PATH variable, with a name that starts with rerun-loader-.

This example is written in Python, and uses TensorFlow to read the files. The events are then logged to Rerun.

NOTE: Not all events are supported yet. Scalars, images, text, and tensors should work. Unsupported events are skipped.

Installing the Rerun Viewer

The simplest option is just:

pip install rerun-sdk

Read this guide for more options.

Installing the plugin

Installing pipx

The most robust way to install the plugin to your PATH is using pipx.

If you don't have pipx installed on your system, you can follow the official instructions here.

Installing the plugin with pipx

Now you can install the plugin to your PATH using

pipx install git+https://github.com/rerun-io/rerun-loader-python-example-tfrecord.git
pipx ensurepath

Note: you can use the --python argument to specify the Python interpreter to use with pipx. On unix-like systems --python $(which python) will use the currently active Python.

Make sure it's installed by running it from your terminal, which should output an error and usage description:

rerun-loader-tfrecord
usage: rerun-loader-tfrecord [-h] [--recording-id RECORDING_ID] filepath
rerun-loader-tfrecord: error: the following arguments are required: filepath

Try it out

Download an example xxx.tfevents.xxx file

curl -OL https://github.com/rerun-io/rerun-loader-python-example-tfrecord/raw/main/events.tfevents.example

Open in the Rerun Viewer

You can either first open the viewer, and then open the file from there using drag-and-drop or the menu>open… dialog, or you can open it directly from the terminal like:

rerun events.tfevents.example