config.json

{
  "inputs": [
    "images"
  ],
  "modules": {
    "attention_module": {
      "config": {
        "args": {
          "depth": 8,
          "maxT": 25,
          "num_channels": 64,
          "scales": [
            [
              32,
              16,
              64
            ],
            [
              128,
              8,
              32
            ],
            [
              512,
              8,
              32
            ]
          ]
        }
      },
      "type": "DeepTextRecognition.CAMModel"
    },
    "feature_extraction": {
      "config": {
        "args": {
          "compress_layer": false,
          "return_multiscale": true,
          "strides": [
            [
              1,
              1
            ],
            [
              2,
              2
            ],
            [
              1,
              1
            ],
            [
              2,
              2
            ],
            [
              1,
              1
            ],
            [
              1,
              1
            ]
          ]
        }
      },
      "type": "DeepTextRecognition.ResNet45Model"
    },
    "processing": {
      "config": {
        "args": {
          "channels_size": 1,
          "image_size": [
            32,
            128
          ],
          "normalize": [
            0.5,
            0.5
          ],
          "padding": "none",
          "resize_method": "bilinear"
        }
      },
      "type": "DeepTextRecognition.ImageProcessor"
    },
    "text_decoder": {
      "config": {
        "args": {
          "dropout": 0.3,
          "nchannel": 512,
          "nclass": 38
        }
      },
      "type": "DeepTextRecognition.DTDModel"
    },
    "tokenizer": {
      "config": {
        "args": {
          "case_sensitive": false,
          "characters": [
            "a",
            "b",
            "c",
            "d",
            "e",
            "f",
            "g",
            "h",
            "i",
            "j",
            "k",
            "l",
            "m",
            "n",
            "o",
            "p",
            "q",
            "r",
            "s",
            "t",
            "u",
            "v",
            "w",
            "x",
            "y",
            "z",
            "1",
            "2",
            "3",
            "4",
            "5",
            "6",
            "7",
            "8",
            "9",
            "0"
          ]
        }
      },
      "type": "DeepTextRecognition.DANTokenizer"
    }
  },
  "order": [
    "processing",
    "feature_extraction",
    "attention_module",
    "text_decoder",
    "tokenizer"
  ],
  "outputs": [
    "tokenizer:labels",
    "tokenizer:probabilities"
  ],
  "routing": {
    "attention_module": {
      "inputs": [
        "feature_extraction:extracted_features"
      ],
      "outputs": [
        "attention_module:attention_features"
      ]
    },
    "feature_extraction": {
      "inputs": [
        "processing:processed_images"
      ],
      "outputs": [
        "feature_extraction:extracted_features"
      ]
    },
    "processing": {
      "inputs": [
        "images"
      ],
      "outputs": [
        "processing:processed_images"
      ]
    },
    "text_decoder": {
      "inputs": [
        "feature_extraction:extracted_features",
        "attention_module:attention_features"
      ],
      "outputs": [
        "text_decoder:predictions",
        "text_decoder:lengths"
      ]
    },
    "tokenizer": {
      "inputs": [
        "text_decoder:predictions",
        "text_decoder:lengths"
      ],
      "outputs": [
        "tokenizer:labels",
        "tokenizer:probabilities"
      ]
    }
  }
}