unet-bob-ross

Handcrafted UNet pytorch training script. Uses the Segmented Bob Ross Images dataset on Kaggle.

To run code and notebooks in a Spell workspace:

spell jupyter --lab \
  --github-url https://github.com/spellml/unet-bob-ross.git \
  --pip kaggle \
  --env KAGGLE_USERNAME=YOUR_USERNAME \
  --env KAGGLE_KEY=YOUR_KEY \
  unet-bob-ross

To execute the training scripts in a Spell run:

spell run \
  --machine-type t4 \
  --github-url https://github.com/spellml/unet-bob-ross.git \
  --pip kaggle \
  --env KAGGLE_USERNAME=YOUR_USERNAME \
  --env KAGGLE_KEY=YOUR_KEY \
  --tensorboard-dir /spell/tensorboards/model_3 \
  "chmod +x /spell/scripts/download_data.sh /spell/scripts/upgrade_env.sh; /spell/scripts/download_data.sh; /spell/scripts/upgrade_env.sh; python /spell/models/model_3.py"

spell run \
  --machine-type t4 \
  --github-url https://github.com/spellml/unet-bob-ross.git \
  --pip transformers --pip tokenizers --pip kaggle \
  --env KAGGLE_USERNAME=YOUR_USERNAME \
  --env KAGGLE_KEY=YOUR_KEY \
  --tensorboard-dir /spell/tensorboards/model_4 \
  "chmod +x /spell/scripts/download_data.sh /spell/scripts/upgrade_env.sh; /spell/scripts/download_data.sh; /spell/scripts/upgrade_env.sh; python /spell/models/model_4.py"