FiftyOne

The FiftyOne app visualizes the predictions of one or multiple trained models, overlaid on labeled frames.

Creating FiftyOne.Dataset for predictions

The first step is to create a FiftyOne.Dataset object (i.e., a Mongo database pointing to images, keypoint predictions, names, and confidences). Run the following command from inside the lightning-pose directory (make sure you have activated your conda environment):

python scripts/create_fiftyone_dataset.py \
--config-path=<PATH/TO/YOUR/CONFIGS/DIR> \
--config-name=<CONFIG_NAME.yaml> \
eval.fiftyone.dataset_name=<YOUR_DATASET_NAME> \
eval.hydra_paths=["</ABSOLUTE/PATH/TO/HYDRA/DIR/1>", "</ABSOLUTE/PATH/TO/HYDRA/DIR/1>"] \
eval.fiftyone.model_display_names=["<NAME_FOR_MODEL_1>","<NAME_FOR_MODEL_2>"] \
eval.fiftyone.launch_app_from_script=true

  • config-path/config-name: these are used the same as the training and inference scripts

  • eval.fiftyone.dataset_name: unique name of FiftyOne.Dataset object

  • eval.hydra_paths: list of absolute paths to directories of the trained models you want to use for prediction. Each directory should contain a predictions.csv file.

    You can also use the relative form

    eval.hydra_paths: ["YYYY-MM-DD/HH-MM-SS/", "YYYY-MM-DD/HH-MM-SS/"]

    which will look in the lightning-pose/outputs directory for these subdirectories.

  • eval.fiftyone.model_display_names: meaningful display names for the models above, e.g.

    eval.fiftyone.model_display_names: ["supervised", "semi-supervised"]

  • eval.fiftyone.launch_app_from_script: if true, the FiftyOne app will launch after dataset creation. To open the app, follow the link provided in the terminal; it should look something like

    http://localhost:5151/

Note

These arguments can also be edited and saved in the config files if needed.

Launching the FiftyOne app from ipython

You can access previously created FiftyOne.Dataset objects from the terminal.

Open an ipython session from your terminal:

ipython

Inside the interactive python session, we import fiftyone, load the dataset we created, and launch the app:

import fiftyone as fo
dataset = fo.load_dataset("<YOUR_DATASET_NAME>")
session = fo.launch_app(dataset)

The app should automatically open in your browser.

If you have forgotten your dataset names, you can list them in ipython:

fo.list_datasets()