semantic_seg.yaml

train:
  experiment_name: 'semantic_sam'

  # Model
  model:
    sam_name: 'sem_sam'
    params:
      # Fix the a part of parameters in SAM
      fix_img_en: True
      fix_prompt_en: True
      fix_mask_de: False
      ckpt_path: 'checkpoints/sam_vit_h_4b8939.pth'
      class_num: 2 # 20 + 1
      model_type: 'vit_h'    # type should be in [vit_h, vit_b, vit_l, default]

  # Dataset
  # dataset:
  #   name: 'torch_voc_sem'
  #   params:
  #     root: '/data/jinziqi/DATASETS/'
  #     year: '2012'
  #     image_set: 'train'
  #   transforms:
  #     resize:
  #       params:
  #         size: [1024, 1024]
  #     to_tensor:
  #       params: ~
  #   target_transforms:
  #     resize:
  #       params:
  #         size: [1024, 1024]


# /path/to/my_dataset
# │
# └───img
# │   └───train
# │   │   │   sample0001.png
# │   │   │   sample0002.png
# │   │   │   ...
# │   └───val
# │       │   sample0003.png
# │       │   sample0004.png
# │       │   ...
# │   
# └───ann
#     └───train
#     │   │   sample0001.png
#     │   │   sample0002.png
#     │   │   ...
#     └───val
#         │   sample0003.png
#         │   sample0004.png
#         │   ...

  # Dataset
  dataset:
    name: 'base_sem'
    params:
      dataset_dir: './data/'
      metainfo:
        class_names: ['background', 'foreground']
      image_set: 'train'
      img_suffix: '.png'
      ann_suffix: '.png'
      data_prefix:
        img_path: 'img'
        ann_path: 'ann'
    transforms:
      resize:
        params:
          size: [1024, 1024]
      to_tensor:
        params: ~
    target_transforms:
      resize:
        params:
          size: [1024, 1024]

  # Losses
  losses:
    ce:
      weight: 0.5
      params:  # ~ means None type, the initial params of loss could be identified here
        ignore_index: 255
      label_one_hot: False

  # Optimizer
  opt_params:
    lr_default:  1e-3
    wd_default: 1e-4
    momentum: 0.9
    lr_list:  [ 1e-2, ]
    group_keys: [ [ 'mask_adapter.decoder_head.output_hypernetworks_mlps', ], ]
    wd_list:  [ 0.0, ]
  opt_name: 'sgd' # 'sgd'
  scheduler_name: 'cosine'

  # Runner
  max_iter: 1000
  log_iter: 1
  eval_iter: 5
  runner_name: 'sem_runner'
  # Dataloader
  bs: 8 # 8
  num_workers: 1
  drop_last: True
  # Logger
  use_tensorboard: True
  tensorboard_folder: './experiment/tensorboard'
  log_folder: './experiment/log'
  model_folder: './experiment/model'

val:
  # Dataset
  dataset:
    name: 'base_sem'
    params:
      dataset_dir: './data/'
      metainfo:
        class_names: ['background', 'foreground']
      image_set: 'val'
      img_suffix: '.png'
      ann_suffix: '.png'
      data_prefix:
        img_path: 'img'
        ann_path: 'ann'
    transforms:
      resize:
        params:
          size: [1024, 1024]
      to_tensor:
        params: ~
    target_transforms:
      resize:
        params:
          size: [1024, 1024]

  bs: 8
  num_workers: 1
  drop_last: True


test:
  need_test: True
  model_path: ./experiment/model/semantic_sam/model.pth
  img_folder: ./data/img/val/
  output_folder: ./experiment/output