Skip to content

Latest commit

 

History

History
executable file
·
51 lines (40 loc) · 3.08 KB

tutorial.md

File metadata and controls

executable file
·
51 lines (40 loc) · 3.08 KB

Introductions to some modules.

  • main.py

    • The entry of the whole project.
    • You can specify some arguments through the command line.
    • Launch the Logger & Configer.
    • Init the model & select the phase of the task you selected.

  • configer.py

    • It is used for the maintenance of all the parameters that passed by the command line and hypes file.
    • The priority of the command line is higher than the priority of the hypes file.
    • Pass message to every corner of the framework.
    • It's very very very important!!!

  • datasets

    • Task Loader for each Task Type, and each method have a corresponding data loader.
    • Transform operations, aug_transform is used for data augmentation that is available to all task type. task-related transforms are only for the specified task type.
    • the generator file in the directory named for dataset is the script that processes the data into the defined format.

  • hypes

    • Each sub project will have a corresponding hype file that contains all the parameters for the project.
    • You could change some parameters through the command line or api for the configer, such as "add_value".

  • methods

    • This directory that contains all the train, val, test & deploy scripts for each project.
    • You could specify the phase of project through command line.
    • It is also available for the programmers to use the class for train, test & deploy.

  • loss

    • This directory that contains all loss for each task type.
    • You could use all the loss through the correponding loss manager.

  • models

    • This directory that contains all models for each task type.
    • You could load pretrained backbone models by simply specify the pretrained model path through the command line.

  • utils

    • utils.layers contains all the packaged layers that are easy to use.
    • utils.tools contained some tool classes that benefit to the whole project.

  • val

    • This directory that contains all the validation scripts for each task.

  • vis

    • This directory that contains all the visualization scripts for each task.
    • Parsers are used to render the label files.
    • Visualizers are used to visualize the results during the trainig, testing, and deploying.

  • imagesite

    • ImageSite is used to look through the image files on remote hosts without mounting or scp.
    • Furthermore, we can also submit command of PytorchCV, check the results of checkpoints, submit the results to evaluation server, and so on.