Add FiftyOne datamodule

[lumurillo]

What does this PR do?

Add a new datamodule that loads FiftyOne datasets. This implementation let MART delegate the data handling to FiftyOne, where the user can do data curation tasks.

This data handling would be agnostic to MART, because at the end here is adapted a torchvision dataset.

I set this implementation to be optional. The user can install the fiftyone dependencies by running:

pip install -e .[fiftyone]

Important:

• To create a dataset in FiftyOne, it's recommended to use the CLI feature: https://docs.voxel51.com/cli/index.html#create-datasets
• To launch the FiftyOne app: https://docs.voxel51.com/cli/index.html#launch-the-app

Type of change

Please check all relevant options.

• Improvement (non-breaking)
• Bug fix (non-breaking)
• New feature (non-breaking)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Testing

Please describe the tests that you ran to verify your changes. Consider listing any relevant details of your test configuration.

• pytest
• CUDA_VISIBLE_DEVICES=0 python -m mart experiment=CIFAR10_CNN_Adv trainer=gpu trainer.precision=16 reports 70% (21 sec/epoch).
• CUDA_VISIBLE_DEVICES=0,1 python -m mart experiment=CIFAR10_CNN_Adv trainer=ddp trainer.precision=16 trainer.devices=2 model.optimizer.lr=0.2 trainer.max_steps=2925 datamodule.ims_per_batch=256 datamodule.world_size=2 reports 70% (14 sec/epoch).

Before submitting

• The title is self-explanatory and the description concisely explains the PR
• My PR does only one thing, instead of bundling different changes together
• I list all the breaking changes introduced by this pull request
• I have commented my code
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have run pre-commit hooks with pre-commit run -a command without errors

Did you have fun?

Make sure you had fun coding 🙃

[mzweilin]

Thanks for bringing the new feature. I was able to install, but I don't have an example to test with.

[mzweilin]

I don't think we want to load zoo datasets here, because it tries to index the dataset every time.

Maybe we should revert the changes, and use the FiftyOne command line to pre-load zoo datasets as datasets. For example, fiftyone zoo datasets load -s validation -n coco-2017-validation coco-2017 loads the validation split of coco-2017 as a dataset named "coco-2017-validation".

[lumurillo]

ok, makes sense

[mzweilin]

FiftyOne commands to load (index) datasets.

Use COCO-2017 as an example. Unfortunately, FiftyOne does not support person-keypoints annotations yet.

Download and load zoo datasets

fiftyone zoo datasets load \
coco-2017 \
-s train \
-n coco-2017-instances-train \
-k include_id=true label_types=detections,segmentations

fiftyone zoo datasets load \
coco-2017 \
-s validation \
-n coco-2017-instances-validation \
-k include_id=true label_types=detections,segmentations

Load local datasets

fiftyone datasets create \
--name coco-2017-instances-validation \
--dataset-dir /raid/datasets/coco/ \
--type fiftyone.types.COCODetectionDataset \
--kwargs \
data_path="val2017" \
labels_path=/raid/datasets/coco/annotations/instances_val2017.json \
persistent=true \
include_id=true

Use the FiftyOne datamodule

datamodule:
  train_dataset:
    dataset_name: coco-2017-instances-train
    gt_field: segmentations
  val_dataset:
    dataset_name: coco-2017-instances-validation
    gt_field: segmentations

[lumurillo]

FiftyOne commands to load (index) datasets.

Use COCO-2017 as an example. Unfortunately, FiftyOne does not support person-keypoints annotations yet.

Download and load zoo datasets

fiftyone zoo datasets load \
coco-2017 \
-s train \
-n coco-2017-instances-train \
-k include_id=true label_types=detections,segmentations

fiftyone zoo datasets load \
coco-2017 \
-s validation \
-n coco-2017-instances-validation \
-k include_id=true label_types=detections,segmentations

Load local datasets

fiftyone datasets create \
--name coco-2017-instances-validation \
--dataset-dir /raid/datasets/coco/ \
--type fiftyone.types.COCODetectionDataset \
--kwargs \
data_path="val2017" \
labels_path=/raid/datasets/coco/annotations/instances_val2017.json \
persistent=true \
include_id=true

Use the FiftyOne datamodule

datamodule:
  train_dataset:
    dataset_name: coco-2017-instances-train
    gt_field: segmentations
  val_dataset:
    dataset_name: coco-2017-instances-validation
    gt_field: segmentations

Is this for the for the example documentation?

[mzweilin]

LGTM!