Install the following:
- Python >= 3.6
- pytorch >= 1.0.1
- torchvision >= 0.2.2
Then, install python packages:
pip install -r requirements.txt
Download Visual Domain Decathlon Datasets(except ImageNet), Caltech256, CIFAR-10, Sketches and move 12 datasets to $DIR/decathlon-1.0-data.
We have rearranged the format of all datasets to suit torchvision.
cd $DIR/decathlon-1.0-data
# 1/10 size
python dataset_shrink.py --save-dir decathlon-1.0-data-tenth --shrink-ratio 10
cd $DIR/src
python decathlon_baseline.py --data-dir decathlon-1.0-data --log-dir log_save --model-save-dir model_weights --depth 28 --widen-factor 1
| Train Dataset | Finetune | Command |
|---|---|---|
| 100% | Fc layer | python decathlon_transfer.py --data-dir decathlon-1.0-data --transfer-result-dir transfer_result_fc_all --fc |
| 100% | All | python decathlon_transfer.py --data-dir decathlon-1.0-data --transfer-result-dir transfer_result_all |
| 10% | Fc layer | python decathlon_transfer.py --data-dir decathlon-1.0-data-tenth --transfer-result-dir transfer_result_fc_all_tenth --fc |
| 10% | All | python decathlon_transfer.py --data-dir decathlon-1.0-data-tenth --transfer-result-dir transfer_result_all_tenth --fc |
We thank Mtan for providing some source codes.