Table of Contents
  • Docs >
  • FiftyOne Environments
Contents

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

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:

1
2
3
4
5
6
# On local machine
import fiftyone as fo

dataset = fo.Dataset(name="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:

1
2
3
4
5
6
# 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.

If you do not have fiftyone installed on your local machine, 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/servce 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.

Notebooks

FiftyOne officialy supports Jupyter Notebooks and Google Colab Notebooks.

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

1
!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.

1
2
3
4
5
6
import fiftyone as fo

dataset = fo.Dataset(name="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 replaced with a screenshot of itself.

An App that was replaced with a screenshot can be reactivated by clicking on the screenshot if within the notebooj environment in which it was created. Note that the reactivated App will load the current state of the session object, not the state in which the screenshot was taken.

1
2
# A new App window will be created in the output of this cell
session.view = dataset.take(10)

A screenshot of the active App can be taken with session.freeze(). This is useful when you are finished with your notebook and ready to share it with others.

1
2
3
# Ensure only screenshots of FiftyOne Apps exist, so the notebook can be
# shared
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():

1
2
# 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():

1
2
3
4
5
6
7
# Update the session's view; no App windows 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 and then call session.open_tab():

1
2
3
# Launch the App in a dedicated browser tab
session = fo.launch_app(dataset, auto=False)
session.open_tab()

Using the desktop App

If you are working from a Jupyter notebook on a machine with the FiftyOne Desktop App installed, you can optionally open the desktop App rather than working with the App in cell output(s).

To do this, pass the desktop=True flag to launch_app():

1
2
# Creates a session and launches the desktop App
session = fo.launch_app(dataset, desktop=True)

Cloud storage

FiftyOne does not yet support accessing data directly in a cloud bucket. Instead, the best practice that we recommend is to mount the cloud bucket as a local drive on a cloud compute instance.

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

Amazon Web Services

If your data is stored in an AWS S3 bucket, we recommend mounting the bucket as a local drive on an EC2 instance and then accessing the data using the standard workflow for remote data.

The steps below outline the process.

Step 1

Create an EC2 instance. We recommend a Linux 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.

We recommend using s3fs-fuse for 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, we recommend mounting the bucket as a local drive on a GC compute instance and then accessing the data using the standard workflow for remote data.

The steps below outline the process.

Step 1

Create a GC compute instance. We recommend a Linux 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.

We recommend using gcsfuse to do this:

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

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, we recommend mounting the bucket as a local drive on an Azure compute instance and then accessing the data using the standard workflow for remote data.

The steps below outline the process.

Step 1

Create an Azure compute instance. We recommend a Linux 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 straight forward if your data is stored in a blob container. We recommend using 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
pip install ipython