{
  "architectures": [
    "VitMatteForImageMatting"
  ],
  "backbone": null,
  "backbone_config": {
    "drop_path_rate": 0.0,
    "dropout_prob": 0.0,
    "hidden_act": "gelu",
    "hidden_size": 2,
    "image_size": 32,
    "initializer_range": 0.02,
    "is_training": true,
    "layer_norm_eps": 1e-06,
    "mlp_ratio": 4,
    "model_type": "vitdet",
    "num_attention_heads": 2,
    "num_channels": 4,
    "num_hidden_layers": 2,
    "out_features": [
      "stage1"
    ],
    "out_indices": [
      1
    ],
    "patch_size": 16,
    "pretrain_image_size": 224,
    "qkv_bias": true,
    "residual_block_indices": [],
    "stage_names": [
      "stem",
      "stage1",
      "stage2"
    ],
    "use_absolute_position_embeddings": true,
    "use_relative_position_embeddings": false,
    "window_block_indices": [],
    "window_size": 0
  },
  "batch_norm_eps": 1e-05,
  "convstream_hidden_sizes": [
    48,
    96,
    192
  ],
  "dtype": "float32",
  "fusion_hidden_sizes": [
    128,
    64,
    32,
    16
  ],
  "hidden_size": 2,
  "initializer_range": 0.02,
  "model_type": "vitmatte",
  "transformers_version": "5.16.0.dev0"
}