image_classifier_3d.utils package¶

Submodules¶

image_classifier_3d.utils.misc_utils.mix_collate(batch)[source]¶: A util function for handling data when images in each mini-batch have different sizes. Conventionally, all images in each mini-batch during training should have the same size. This is a work-around. Inspired by https://discuss.pytorch.org/t/how-to-create-a-dataloader-with-variable-size-input/8278/3 # noqa E501

class image_classifier_3d.utils.quilt_utils.QuiltModelZoo[source]¶

Bases: object

connect to model zoo on quilt3

download_model(model_name: str, save_path: Union[str, pathlib.Path] = './zoo/model.pth')[source]¶

download the model “model_name” to “out_path”

Paremeters — model_name: str

the name of the model to be downloaded

save_path: Union(str, Path): the path to save the model, default is ‘./zoo/model.pth’

peak_all_models() → List[source]¶: print out names of existing models in the model zoo

image_classifier_3d.utils.quilt_utils.validate_model(config, hparams, save_path)[source]¶

validate the model configuration parameters, for model path, model files will be downloaded from quilt if not exist

config:: a dictionary of key parameters
hparams:: a Namespace object similar to config (currently redudent, will be improved in future versions)
save_path:: if the required models are not found in the save_path, model files will be downloaded to save_path