FiftyOne Environments

This guide describes best practices for using FiftyOne with data stored in various environments, including local machines, remote servers, and cloud storage.

Terminology

• Local machine: Data is stored on the same computer that will be used to launch the App

• Remote machine: Data is stored on disk on a separate machine (typically a remote server) from the one that will be used to launch the App

• Notebooks: You are working from a Jupyter Notebook, Google Colab Notebook, or Databricks Notebook SageMaker Notebook

• Cloud storage: Data is stored in a cloud bucket (e.g., S3, GCS, or Azure)

Local data

When working with data that is stored on disk on a machine with a display, you can directly load a dataset and then launch the App:

# On local machine
import fiftyone as fo

dataset = fo.Dataset("my-dataset")

session = fo.launch_app(dataset)  # (optional) port=XXXX

From here, you can explore the dataset interactively from the App and from your Python shell by manipulating the session object.

Note

You can use custom ports when launching the App in order to operate multiple App instances simultaneously on your machine.

Remote data

FiftyOne supports working with data that is stored on a remote machine that you have ssh access to. The basic workflow is to load a dataset on the remote machine via the FiftyOne Python library, launch a remote session, and connect to the session on your local machine where you can then interact with the App.

First, ssh into your remote machine and install FiftyOne if necessary.

Then load a dataset using Python on the remote machine and launch a remote session:

# On remote machine
import fiftyone as fo

dataset = fo.load_dataset(...)

session = fo.launch_app(dataset, remote=True)  # optional: port=XXXX

Leave the Python REPL running and follow the instructions for connecting to this session remotely that were printed to your terminal (also described below).

Note

You can manipulate the session object on the remote machine as usual to programmatically interact with the App instance that you view locally.

To connect to your remote session, open a new terminal window on your local machine and execute the following command to setup port forwarding to connect to your remote session:

# On local machine
ssh -N -L 5151:127.0.0.1:XXXX [<username>@]<hostname>

Leave this process running and open http://localhost:5151 in your browser to access the App.

In the above, [<username>@]<hostname> specifies the remote machine to connect to, XXXX refers to the port that you chose when you launched the session on your remote machine (the default is 5151), and 5151 specifies the local port to use to connect to the App (and can be customized).

Alternatively, if you have FiftyOne installed on your local machine, you can use the CLI to automatically configure port forwarding and open the App in your browser as follows:

# On local machine
fiftyone app connect --destination [<username>@]<hostname>

If you choose a custom port XXXX on the remote machine, add a --port XXXX flag to the above command.

If you would like to use a custom local port, add a --local-port YYYY flag to the above command.

Note

You can customize the local/remote ports used when launching remote sessions in order to connect/service multiple remote sessions simultaneously.

Note

If you use ssh keys to connect to your remote machine, you can use the optional --ssh-key argument of the fiftyone app connect command.

However, if you are using this key regularly, it is recommended to add it to your ~/.ssh/config as the default IdentityFile.

Restricting the App address

By default, the App will listen on localhost. However, you can provide the optional address parameter to launch_app() to specify a particular IP address or hostname for the App to listen on.

Using the default of localhost means the App can only be accessed from the local machine or a machine that was able to setup ssh port forwarding as described in the previous section.

An alternative is to set the App address to "0.0.0.0" so that the App can be accessed from a remote host or from the local machine itself. Using "0.0.0.0" will bind the App to all available interfaces and will allow access to the App from any remote resource with access to your network.

import fiftyone as fo

dataset = fo.load_dataset(...)

# Enable connections from remote hosts
session = fo.launch_app(dataset, remote=True, address="0.0.0.0")

If desired, you can permanently configure an App address by setting the default_app_address of your FiftyOne config. You can achieve this by adding the following entry to your ~/.fiftyone/config.json file:

{
    "default_app_address": "0.0.0.0"
}

or by setting the following environment variable:

export FIFTYONE_DEFAULT_APP_ADDRESS='0.0.0.0'

Notebooks

FiftyOne officially supports Jupyter Notebooks, Google Colab Notebooks, Databricks Notebooks. App support is also available in SageMaker Notebooks and any cloud notebook that has an accessible network proxy via configured proxy_url.

To use FiftyOne in a notebook, simply install fiftyone via pip:

!pip install fiftyone

and load datasets as usual. When you run launch_app() in a notebook, an App window will be opened in the output of your current cell.

import fiftyone as fo

dataset = fo.Dataset("my-dataset")

# Creates a session and opens the App in the output of the cell
session = fo.launch_app(dataset)

Any time you update the state of your session object; e.g., by setting session.dataset or session.view, a new App window will be automatically opened in the output of the current cell. The previously active App will be “frozen”, i.e., replaced with a screenshot of its current state.

# A new App window will be created in the output of this cell, and the
# previously active App instance will be replaced with a screenshot
session.view = dataset.take(10)

You can reactivate a frozen App instance from the same notebook session by clicking on the screenshot.

Note

Reactivating a frozen App instance will load the current state of the session object, not the state in which the screenshot was taken.

To reactivate an App instance from a previous session, e.g., when running a notebook downloaded from the web for the first time, you must (re)run the cell.

You can manually replace the active App instance with a screenshot by calling session.freeze(). This is useful when you are finished with your notebook and ready to share it with others, as an active App instance itself cannot be viewed outside of the current notebook session.

# Replace active App instance with screenshot so App state is viewable offline
session.freeze()

Manually controlling App instances

If you would like to manually control when new App instances are created in a notebook, you can pass the auto=False flag to launch_app():

# Creates a session but does not open an App instance
session = fo.launch_app(dataset, auto=False)

When auto=False is provided, a new App window is created only when you call session.show():

# Update the session's view; no App window is created
session.view = dataset.take(10)

# In another cell

# Now open an App window in the cell's output
session.show()

As usual, this App window will remain connected to your session object, so it will stay in-sync with your session whenever it is active.

Note

If you run session.show() in multiple cells, only the most recently created App window will be active, i.e., synced with the session object.

You can reactivate an older cell by clicking the link in the deactivated App window, or by running the cell again. This will deactivate the previously active cell.

Opening the App in a dedicated tab

If you are working from a Jupyter notebook, you can open the App in a separate browser tab rather than working with it in cell output(s).

To do this, pass the auto=False flag to launch_app() when you launch the App (so that additional App instances will not be created as you work) and then call session.open_tab():

# Launch the App in a dedicated browser tab
session = fo.launch_app(dataset, auto=False)
session.open_tab()

Remote notebooks

You can also work in a Jupyter notebook in your local browser that is served from a remote machine where your data is located. Follow the instructions below to achieve this.

On the remote machine:

Start the Jupyter server on a port of your choice:

# On remote machine
jupyter notebook --no-browser --port=XXXX /path/to/notebook.ipynb

On your local machine:

Back on your local machine, you will need to forward the remote port XXXX to a local port (we’ll also use XXXX here, for consistency):

# On local machine
ssh -N -L XXXX:localhost:XXXX [<username>@]<hostname>

Now open localhost:XXXX in your browser and you should find your notebook!

If your notebook launches the FiftyOne App, you will also need to forward the port used by the App to your local machine. By default, the App uses port 5151, but you can specify any port, say YYYY, not currently in use on your remote machine:

# On local machine
ssh -N -L 5151:localhost:YYYY [<username>@]<hostname>

In your Jupyter notebook:

When you launch the FiftyOne App in your notebook, you should now see the App as expected!

# Launch the App in a notebook cell
session = fo.launch_app(dataset)  # port=YYYY

If you chose a port YYYY other than the default 5151, you will need to specify it when launching App instances per the commented argument above.

Note that you can also open the App in a dedicated tab:

# Launch the App in a dedicated browser tab
session = fo.launch_app(dataset, auto=False)  # port=YYYY
session.open_tab()

Docker

The FiftyOne repository contains a Dockerfile that you can use/customize to build and run Docker images containing source or release builds of FiftyOne.

Building an image

First, clone the repository:

git clone https://github.com/voxel51/fiftyone
cd fiftyone

If you want a source install of FiftyOne, then build a wheel:

make python

If you want to install a FiftyOne release, then make the suggested modification in the Dockerfile.

Next, build the image:

docker build -t voxel51/fiftyone .

The default image uses Python 3.11, but you can customize these via optional build arguments:

docker build \
    --build-arg PYTHON_VERSION=3.10 \
    -t voxel51/fiftyone .

Refer to the Dockerfile for additional Python packages that you may wish to include in your build.

Running an image

The image is designed to persist all data in a single /fiftyone directory with the following organization:

/fiftyone/
    db/             # FIFTYONE_DATABASE_DIR
    default/        # FIFTYONE_DEFAULT_DATASET_DIR
    zoo/
        datasets/   # FIFTYONE_DATASET_ZOO_DIR
        models/     # FIFTYONE_MODEL_ZOO_DIR

Therefore, to run a container, you should mount /fiftyone as a local volume via --mount or -v, as shown below:

SHARED_DIR=/path/to/shared/dir

docker run -v ${SHARED_DIR}:/fiftyone -p 5151:5151 -it voxel51/fiftyone

The -p 5151:5151 option is required so that when you launch the App from within the container you can connect to it at http://localhost:5151 in your browser.

You can also include the -e or --env-file options if you need to further configure FiftyOne.

By default, running the image launches an IPython shell, which you can use as normal:

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart")
session = fo.launch_app(dataset)

Note

Any datasets you create inside the Docker image must refer to media files within SHARED_DIR or another mounted volume if you intend to work with datasets between sessions.

Note

FiftyOne should automatically detect that it is running inside a Docker container. However, if you are unable to load the App in your browser, you may need to manually set the App address to 0.0.0.0:

session = fo.launch_app(..., address="0.0.0.0")

Connecting to a localhost database

If you are using a self-managed database that you ordinarily connect to via a URI like mongodb://localhost, then you will need to tweak this slightly when working in Docker. See this question for details.

On Linux, include --network="host" in your docker run command and use mongodb://127.0.0.1 for your URI.

On Mac or Windows, use mongodb://host.docker.internal for your URI.

Cloud storage

For prototyping, it is possible to work with data in cloud storage buckets in FiftyOne by mounting the buckets as local drives.

The following sections describe how to do this in the AWS, Google Cloud, and Microsoft Azure environments.

Warning

Mounting cloud buckets using the techniques below is not performant and is not recommended or officially supported. It is useful only for prototyping.

Our recommended, scalable approach to work with cloud-backed data is FiftyOne Teams, an enterprise deployment of FiftyOne with multiuser collaboration features, native cloud dataset support, and much more!

AWS

If your data is stored in an AWS S3 bucket, you can mount the bucket as a local drive on an EC2 instance and then access the data using the standard workflow for remote data.

The steps below outline the process.

Step 1

Create an EC2 instance.

Step 2

Now ssh into the instance and install FiftyOne if necessary.

# On remote machine
pip install fiftyone

Note

You may need to install some system packages on your compute instance instance in order to run FiftyOne.

Step 3

Mount the S3 bucket as a local drive.

You can use s3fs-fuse to do this. You will need to make a .passwd-s3fs file that contains your AWS credentials as outlined in the s3fs-fuse README.

# On remote machine
s3fs <bucket-name> /path/to/mount/point \
    -o passwd_file=.passwd-s3fs \
    -o umask=0007,uid=<your-user-id>

Step 4

Now that you can access your data from the compute instance, start up Python and create a FiftyOne dataset whose filepaths are in the mount point you specified above. Then you can launch the App and work with it locally in your browser using remote sessions.

Google Cloud

If your data is stored in a Google Cloud storage bucket, you can mount the bucket as a local drive on a GC compute instance and then access the data using the standard workflow for remote data.

The steps below outline the process.

Step 1

Create a GC compute instance.

Step 2

Now ssh into the instance and install FiftyOne if necessary.

# On remote machine
pip install fiftyone

Note

You may need to install some system packages on your compute instance instance in order to run FiftyOne.

Step 3

Mount the GCS bucket as a local drive.

You can use gcsfuse to do this:

# On remote machine
gcsfuse --implicit-dirs my-bucket /path/to/mount

Step 4

Now that you can access your data from the compute instance, start up Python and create a FiftyOne dataset whose filepaths are in the mount point you specified above. Then you can launch the App and work with it locally in your browser using remote sessions.

Microsoft Azure

If your data is stored in an Azure storage bucket, you can mount the bucket as a local drive on an Azure compute instance and then access the data using the standard workflow for remote data.

The steps below outline the process.

Step 1

Create an Azure compute instance.

Step 2

Now ssh into the instance and install FiftyOne if necessary.

# On remote machine
pip install fiftyone

Note

You may need to install some system packages on your compute instance instance in order to run FiftyOne.

Step 3

Mount the Azure storage container in the instance.

This is fairly straightforward if your data is stored in a blob container. You can use blobfuse for this.

Step 4

Now that you can access your data from the compute instance, start up Python and create a FiftyOne dataset whose filepaths are in the mount point you specified above. Then you can launch the App and work with it locally in your browser using remote sessions.

Setting up a cloud instance

When you create a fresh cloud compute instance, you may need to install some system packages in order to install and use FiftyOne.

For example, the script below shows a set of commands that may be used to configure a Debian-like Linux instance, after which you should be able to successfully install FiftyOne.

# Example setup script for a Debian-like virtual machine

# System packages
sudo apt update
sudo apt -y upgrade
sudo apt install -y build-essential
sudo apt install -y unzip
sudo apt install -y cmake
sudo apt install -y cmake-data
sudo apt install -y pkg-config
sudo apt install -y libsm6
sudo apt install -y libxext6
sudo apt install -y libssl-dev
sudo apt install -y libffi-dev
sudo apt install -y libxml2-dev
sudo apt install -y libxslt1-dev
sudo apt install -y zlib1g-dev
sudo apt install -y python3
sudo apt install -y python-dev
sudo apt install -y python3-dev
sudo apt install -y python3-pip
sudo apt install -y python3-venv
sudo apt install -y ffmpeg  # if working with video

# (Recommended) Create a virtual environment
python3 -m venv fiftyone-env
. fiftyone-env/bin/activate

# Python packages
pip install --upgrade pip setuptools wheel build
pip install ipython