Skip to content

kaylode/theseus

Repository files navigation

⛵ Ship of Theseus

The ship wherein Theseus and the youth of Athens returned from Crete had thirty oars, and was preserved by the Athenians down even to the time of Demetrius Phalereus, for they took away the old planks as they decayed, putting in new and stronger timber in their places, insomuch that this ship became a standing example among the philosophers, for the logical question of things that grow; one side holding that the ship remained the same, and the other contending that it was not the same. — Plutarch, Theseus


📝 Instructions

Installation

  • Create virtual environment: conda create -n myenv python=3.10
  • Install Pytorch. Currently support torch==1.13.1
  • Inside your project, install this package by git+https://github.com/kaylode/theseus.git@master#egg=theseus[cv,cv.classification,cv.detection,cv.semantic] extra packages can be identified from the pyproject.toml.

To adapt for personal project

  1. Create your own dataset, dataloader, model, loss function, metric function, ... and register it to the registry so that it can be generated from config at runtime.
  2. Customize inherited trainer and pipeline to your need, such as what to do before/after training/validating step,...
  3. Write custom callbacks (recommended!), follow Lightning
  4. Modify configuration file

See theseus/cv/classification for example

To execute scripts with arguments

  • Run the script with --config-dir flag with a specified config folder that contains the yaml file. And --config-name is that file's name.

Example:

python train.py \
    --config-dir configs \
    --config-name pipeline.yaml
  • To override arguments inside the .yaml file, follow the instructions from Hydra. For example, to train 50 epochs and resume training from checkpoints:
python train.py \
    --config-dir configs \
    --config-name pipeline.yaml \
    trainer.args.max_epochs=5000 \
    global.resume=checkpoint.pth

Notice: There are no spaces between keys and values in -o flag

🎒 Resources

  • Example colab notebooks for classification tasks: Notebook
  • Example colab notebooks for semantic segmentation tasks: Notebook

📘 References