task_attributes:
  segmentation:
    dataset_attributes:
      few_shot_flood_segmentation:
        data_root_dir: 'eval_datasets/flood3i'
        data_txt: 'eval_datasets/flood3i/val.txt'
        img_dir: 'eval_datasets/flood3i/Images'
        tgt_dir: 'eval_datasets/flood3i/Semantic_mask'
        num_shot: 1
        seq_len: 1
        npz_key: 'arr_0'
        image_size:
          hr: (512, 512)
          s2: (16, 16)
          s1: (16, 16)
          anno: (512, 512)
        mim:
          input_size: (1024, 512)
          patch_size: 128
          mask_ratio: 0.5

model_attributes:
  SkySensePP:
    sources: ['hr', 's2', 's1']
    use_glbank: False
    use_modal_vae: True
    use_ctpe: False
    use_cls_token_uper_head: False 
    upsacle_results: True
    calendar_time: 365
    vocabulary_size: 64
    backbone_hr:
      type: 'SwinTransformerV2MSL'
      arch: 'huge'
      use_attn: True
      merge_stage: 2
      vocabulary_size: 64
      img_size: 224
      patch_size: 4
      in_channels: 3
      window_size: 8
      drop_rate: 0.
      drop_path_rate: 0.2
      out_indices: (0,1,2,3)
      use_abs_pos_embed: False
      interpolate_mode: 'bicubic'
      with_cp: True
      frozen_stages: -1
      norm_eval: False
      pad_small_map: False
      pretrained_window_sizes: [0, 0, 0, 0]

    backbone_s2:
      type: 'VisionTransformerMSL'
      img_size: (16, 16)
      use_attn: False
      merge_stage: 4
      vocabulary_size: 64
      patch_size: 4
      in_channels: 10
      embed_dims: 1024
      num_layers: 24
      num_heads: 16
      mlp_ratio: 4
      out_indices: (5,11,17,23)
      qkv_bias: True
      drop_rate: 0.
      attn_drop_rate: 0.
      drop_path_rate: 0.3
      with_cls_token: False
      output_cls_token: False
      act_cfg:
        type: 'GELU'
      norm_cfg:
        type: 'LN'
        eps: 1e-6
      with_cp: True
      interpolate_mode: 'bicubic'
    
    head_s2:
      type: 'UPHead'
      in_dim: 1024
      out_dim: 2816
      up_scale: 4
    
    backbone_s1:
      type: 'VisionTransformerMSL'
      img_size: (16, 16)
      use_attn: False
      merge_stage: 4
      vocabulary_size: 64
      patch_size: 4
      in_channels: 2
      embed_dims: 1024
      num_layers: 24
      num_heads: 16
      mlp_ratio: 4
      out_indices: (5,11,17,23)
      qkv_bias: True
      drop_rate: 0.
      attn_drop_rate: 0.
      drop_path_rate: 0.3
      with_cls_token: False
      output_cls_token: False
      act_cfg:
        type: 'GELU'
      norm_cfg:
        type: 'LN'
        eps: 1e-6
      with_cp: True
      interpolate_mode: 'bicubic'
    
    head_s1:
      type: 'UPHead'
      in_dim: 1024
      out_dim: 2816
      up_scale: 4

    rec_head_hr:
      type: 'UPerHead'
      in_channels: [704, 704, 1408, 2816, 1024]
      in_index: [0, 1, 2, 3, 4]
      pool_scales: (1, 2, 3, 6)
      channels: 512
      dropout_ratio: 0.1
      num_classes: 65
      norm_cfg:
        type: 'SyncBN'
        requires_grad: true
      align_corners: false
    
    necks:
      type: 'TransformerEncoder'
      input_dims: 2816
      embed_dims: 1024
      num_layers: 24
      num_heads: 16
      mlp_ratio: 4
      qkv_bias: True
      drop_rate: 0.
      attn_drop_rate: 0.
      drop_path_rate: 0.3
      with_cls_token: True
      output_cls_token: True
      norm_cfg:
        type: 'LN'
      act_cfg:
        type: 'GELU'
      num_fcs: 2
      norm_eval: False
      with_cp: True

    modality_vae:
      type: 'ModalityCompletion'
      input_shape_hr: [2816, 32, 16]
      input_shape_s2: [2816, 32, 16]
      input_shape_s1: [2816, 32, 16]
      conv_dim: 256
      z_dim: 256
      n_codebook: 8192

amp_attributes:
  amp_escapes: Conv2d
  opt_level: O1
  init_scale: 1

predictor_parameters:
  predictor: 'OneshotPredictor'
  replace_speedup_op: True
  device: cuda
  local_rank: 0