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

# Shelfnet34_lw recipe for COCO segmentation 21 classes from PASCAL.
# Reaches ~65.1 mIOU
# Trained using 4 X 2080 Ti using DDP- takes ~ 2d 7h with batch size of 8 and batch accumulate of 3 (i.e effective batch
# size is 4*8*3 = 96)
# Logs and tensorboards: s3://deci-pretrained-models/shelfnet34_coco_segmentation_tensorboard/


# 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:
#       python src/super_gradients/examples/train_from_recipe_example/train_from_recipe.py --config-name=coco_segmentation_shelfnet_lw


# /!\ THIS RECIPE IS NOT SUPPORTED AT THE MOMENT /!\

defaults:
  - training_hyperparams: coco_segmentation_shelfnet_lw_train_params
  - dataset_params: coco_segmentation_dataset_params
  - arch_params: shelfnet34_lw_arch_params
  - checkpoint_params: default_checkpoint_params
  - _self_

train_dataloader: coco_segmentation_train
val_dataloader: coco_segmentation_val

checkpoint_params:
  strict_load: True
  load_backbone: True
  checkpoint_path:

resume: False
training_hyperparams:
  resume: ${resume}

experiment_name: coco_segmentation_21_subclass_shelfnet34

multi_gpu: DDP
num_gpus: 4

ckpt_root_dir:

architecture: shelfnet34_lw


# 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}}