LTAN-Ligntweight Task Adaptation Network for Cross-Domain Few-Shot Learning

This is the competition code for the NeurIPS 2022 Cross-Domain MetaDL Challenge.

Requirements.

Please follow the official website to get the starting kit, download the public data and set up the environment.

The code was tested with Python 3.7 and Pytorch >= 1.10.0.

Running code under the competition setting

The details of meta-learner, learner, and predictor can be found in model.py. Please follow the official tutorial to run the codes.

Here is an example. First, enter into the cd-metadl folder, and move the package of cdml22-ltan into the baselines folder. Then run the following command to train the model and get test results:

cd path/to/cd-metadl
python -m cdmetadl.run --submission_dir=path/to/this/folder --output_dir_ingestion=ingestion_output --output_dir_scoring=scoring_output --verbose=False  --test_tasks_per_dataset=10

Before running these scripts, please set the dataset directory and model directory in the arguments or directly change the default directories in the code.