Running on your own data

To run on your own data, you will need:

A directory containing your images.
COCO-formatted annotations for your images. Note that COCO bounding box format is [x_min, y_min, bbox_width, bbox_height] (this is not specified in the docs).

You will need a separate COCO file for each dataset. Typically this will include:

Train: Labeled source-domain images for supervised burn-in and training
Unlabeled: Target-domain images for DAOD. These can be optionally labeled if you also want to use them to train a supervised "oracle" for comparison. If unlabeled, the "annotations" entry of your COCO file can be an empty list.
Test: A labeled validation set. In most DAOD papers this comes from your target domain, even though this breaks the constraints of UDA.

To use your dataset with ALDI,

Make your data (images and annotations) accessible from the datasets directory in this repository. You can either place your data here, or add symlinks to it.
Register each of your dataset(s) by adding a new line to datasets.py. Paths should be relative to the base directory of this repo. Note that the images directories can be the same if you'd like, since images will be chosen according to their paths in the COCO JSON files.

register_coco_instances("your_training_dataset_name", {}, "path/to/your_training_coco_labels.json", "path/to/your/train/images/")
register_coco_instances("your_unlabeled_dataset_name", {}, "path/to/your_unlabeled_coco_labels.json", "path/to/your/unlabeled/images/")
register_coco_instances("your_test_dataset_name", {}, "path/to/your_test_coco_labels.json", "path/to/your/test/images/")

Now your datasets can be referenced by name in config.yaml files, e.g.:

DATASETS:
  TRAIN: ("your_training_dataset_name",) # needs to be a tuple, and can contain multiple datasets if you want
  UNLABELED: ("your_unlabeled_dataset_name",) # needs to be a tuple, and can contain multiple datasets if you want
  TEST: ("your_test_dataset_name",)  # needs to be a tuple, and can contain multiple datasets if you want

You will also need to update the MODEL.ROI_HEADS.NUM_CLASSES value in your config files to match the number of classes in your COCO files, e.g.:

MODEL:
  ROI_HEADS:
    NUM_CLASSES: 9 # change to match your number of classes