wbia-orientation 4.0.1

Contributing

It’s recommended that you use pre-commit to ensure linting procedures are run on any commit you make. (See also pre-commit.com)

Reference pre-commit’s installation instructions for software installation on your OS/platform. After you have the software installed, run pre-commit install on the command line. Now every time you commit to this project’s code base the linter procedures will automatically run over the changed files. To run pre-commit on files preemtively from the command line use:

git add .
pre-commit run

# or

pre-commit run --all-files

Brunette

Our code base has been formatted by Brunette, which is a fork and more configurable version of Black (https://black.readthedocs.io/en/stable/).

Flake8

Try to conform to PEP8. You should set up your preferred editor to use flake8 as its Python linter, but pre-commit will ensure compliance before a git commit is completed.

To run flake8 from the command line use:

flake8

This will use the flake8 configuration within setup.cfg, which ignores several errors and stylistic considerations. See the setup.cfg file for a full and accurate listing of stylistic codes to ignore.

PyTest

Our code uses Google-style documentation tests (doctests) that uses pytest and xdoctest to enable full support. To run the tests from the command line use:

pytest

To run doctests with +REQUIRES(–web-tests) do:

pytest --web-tests

Dependencies

• Python >= 3.7

• PyTorch >= 1.5

Data

Data used for training and evaluation:

• sea turtle head parts - orientation.seaturtle.coco.tar.gz

• sea dragon head parts - orientation.seadragon.coco.tar.gz

• manta ray body annotations - orientation.mantaray.coco.tar.gz

• spotted dolphin body annotations - orientation.spotteddolphin.coco.tar.gz

• hammerhead shark body annotations - orientation.hammerhead.coco.tar.gz

• right whale bonnet parts - orientation.rightwhale.coco.tar.gz

• whale shark - orientation.whaleshark.coco.tar.gz

Data preprocessing

Each dataset is preprocessed to speed-up image loading during training. At the first time of running a training or a testing script on a dataset the following operations are applied:

• an object is cropped based on a segmentation boudnding box from annotations with a padding around equal to the half size of the box to allow for image augmentations

• an image is resized so the smaller side is equal to the double size of a model input; the aspect ratio is preserved.

The preprocessed dataset is saved in data directory.

Data augmentations

During the training the data is augmented online in the following way:

• Random Horizontal Flips

• Random Vertical Flips

• Random Rotations

• Random Scale

• Random Crop

• Color Jitter (variations in brightness, hue, contrast and saturation)

Both training and testing data are resized to the model input size and normalized.

Training

Run the training script:

python wbia_orientation/train.py --cfg <path_to_config_file> <additional_optional_params>

Configuration files are listed in experiments folder. For example, the following line trains the model with parameters specified in the config file:

python wbia_orientation/train.py --cfg wbia_orientation/config/mantaray.yaml

To override a parameter in config, add this parameter as a command line argument:

python wbia_orientation/train.py --cfg wbia_orientation/config/mantaray.yaml TRAIN.BS 64

Testing

The test script evaluates on the test set with the best model saved during training:

python wbia_orientation/test.py --cfg <path_to_config_file> <additional_optional_params>

For example:

python wbia_orientation/test.py --cfg wbia_orientation/config/mantaray.yaml

By default, the accuracy of detected rotation angle is computed for a threshold of 10 degrees. Pass a different value as a command line parameter to evaluate with another threshold:

python wbia_orientation/test.py --cfg wbia_orientation/config/mantaray.yaml TEST.THETA_THR 15.

pytest