super-gradients/src/super_gradients/recipes/supervisely_unet.yaml

# Binary segmentation training example of UNet model on the Supervisely person dataset.

# Instructions:
#   0. Make sure that the data is stored in dataset_params.dataset_dir or add "dataset_params.data_dir=<PATH-TO-DATASET>" at the end of the command below (feel free to check ReadMe)
#   1. Move to the project root (where you will find the ReadMe and src folder)
#   2. Run the command:
#      UNet:        python src/super_gradients/examples/train_from_recipe_example/train_from_recipe.py --config-name=supervisely_unet
#
# Validation Target (Person class) IoU and training time:
#      UNet:        input-size: [480, 320]     mIoU: 89.18     1 X RTX A5000, 4 H
#
# Logs, tensorboards and network checkpoints:
#      UNet:       https://deci-pretrained-models.s3.amazonaws.com/unet/supervisely/
#

defaults:
  - training_hyperparams: supervisely_default_train_params
  - dataset_params: supervisely_persons_dataset_params
  - checkpoint_params: default_checkpoint_params
  - _self_

architecture: unet
arch_params:
  num_classes: 1
  use_aux_heads: False

training_hyperparams:
  initial_lr: 0.025

  loss:
    bce_dice_loss:
      loss_weights: [ 1., 1. ]
      logits: True

dataset_params:
  batch_size: 16

multi_gpu: OFF

experiment_name: unet_supervisely
ckpt_root_dir:


# THE FOLLOWING PARAMS ARE DIRECTLY USED BY HYDRA
hydra:
  run:
    # Set the output directory (i.e. where .hydra folder that logs all the input params will be generated)
    dir: ${hydra_output_dir:${ckpt_root_dir}, ${experiment_name}}