package tensorflow_hub

Get desktop application:
View/edit binary Protocol Buffers messages

Internal representation of properties of image modules. A TensorFlow Hub module that operates on images can have an attached_message "image_module_info" of this type to provide image-related information to the the code in image_util.py. Module consumers should use those library functions; the details of this protocol message are subject to change.

• optional ImageModuleInfo.InputSize default_image_size = 1

  The size information for all image inputs in all signatures.

Size of input images. A module that accepts an image input (or multiple ones) can describe its expectations on the image size in this message

Used in: ImageModuleInfo

• int32 height = 1

  The expected input size (in pixels) that the module was trained for. This can be reported here without constraining the input's shape in the signature, in case the module can also handle other sizes.

• int32 width = 2

Internal representation of attached messages in FORMAT_V3 of NativeModules. Each graph in the SavedModel contains a CollectionDef with ModuleAttachments serialized as a BytesList. For more information and the public API, see ModuleSpec.get_attached_message().

• string key = 1

  The unique key by which the attached message is stored and retrieved.

• bytes value = 2

  The attached message in serialized form.