Node attributes in deserialised JSON contain strings as lists of ascii codes. The TF GraphDef proto expects these values to be base64 encoded

Convert JSON to TF GraphDef message Args: message_dict: deserialised JSON message Returns: TF GraphDef message

Convert TFJS JSON model to TF Graph Args: model_json: JSON dict from TFJS model file base_path: Path to the model file (wher

Deep search string attributes (labelled "s" in GraphDef proto) and convert ascii code lists to base64-encoded strings if necessary

Convert list of weight entries to dictionary Args: weight_list: List of numpy arrays or tensors formatted as {'n

Create a TF Graph from nodes Args: graph_def: TF GraphDef message containing the node graph weight_dict: Dictionary from node

Recursively find all objects with a given key in a dictionary Args: obj: Dictionary to search key: Key to find of_typ

Search dilations-attribute and convert misaligned dilation rates if necessary see https://github.com/patlevin/tfjs-to-tf/issues/1

Parse model directory name and return path and file name Args: model_dir: Model file path - either directory name or path + file name

Return information about a graph's inputs. Arguments: graph: Graph or GraphDef object Returns: List of NodeInfo objects h

Return the names of the graph's input tensors. Arguments: graph: Graph or GraphDef object Returns: List of tensor names

Return information about a graph's outputs. Arguments: graph: Graph or GraphDef object Returns: List of NodeInfo objects

Return the names of the graph's output tensors. Arguments: graph: Graph or GraphDef object Returns: List of tensor names

Load a TFJS Graph Model from a directory Args: model_dir: Directory that contains the tfjs model.json and weights; al