{
  "architectures": [
    "SamHQModel"
  ],
  "dtype": "float32",
  "initializer_range": 0.02,
  "mask_decoder_config": {
    "attention_downsample_rate": 2,
    "hidden_act": "relu",
    "hidden_size": 32,
    "iou_head_depth": 3,
    "iou_head_hidden_dim": 32,
    "layer_norm_eps": 1e-06,
    "mlp_dim": 64,
    "model_type": "",
    "num_attention_heads": 4,
    "num_hidden_layers": 12,
    "num_multimask_outputs": 3,
    "vit_dim": 36
  },
  "model_type": "sam_hq",
  "prompt_encoder_config": {
    "hidden_act": "gelu",
    "hidden_size": 32,
    "image_embedding_size": 12,
    "image_size": 24,
    "layer_norm_eps": 1e-06,
    "mask_input_channels": 4,
    "model_type": "",
    "num_point_embeddings": 4,
    "patch_size": 2
  },
  "tie_word_embeddings": true,
  "transformers_version": "5.16.0.dev0",
  "vision_config": {
    "attention_dropout": 0.0,
    "dropout": 0.0,
    "global_attn_indexes": [
      2,
      5,
      8,
      11
    ],
    "hidden_act": "gelu",
    "hidden_size": 36,
    "image_size": 24,
    "initializer_factor": 1.0,
    "initializer_range": 0.02,
    "intermediate_size": 72,
    "layer_norm_eps": 1e-06,
    "mlp_dim": 144,
    "mlp_ratio": 4.0,
    "model_type": "sam_hq_vision_model",
    "num_attention_heads": 4,
    "num_channels": 3,
    "num_hidden_layers": 12,
    "num_pos_feats": 16,
    "output_channels": 32,
    "patch_size": 2,
    "projection_dim": 62,
    "qkv_bias": true,
    "rel_pos_zero_init": false,
    "scale": 18,
    "use_abs_pos": true,
    "use_rel_pos": true,
    "window_size": 14
  }
}