Python caffe2.proto.caffe2_pb2.DeviceOption() Examples

def _tf_device(device_option):
    '''
    Handle the devices.

    Args:
        device_option (caffe2_pb2.DeviceOption): DeviceOption protobuf,
            associated to an operator, that contains information such as
            device_type (optional), cuda_gpu_id (optional), node_name (optional,
            tells which node the operator should execute on). See caffe2.proto
            in caffe2/proto for the full list.

    Returns:
        Formatted string representing device information contained in
            device_option.
    '''
    if not device_option.HasField("device_type"):
        return ""
    if device_option.device_type == caffe2_pb2.CPU or device_option.device_type == caffe2_pb2.MKLDNN:
        return "/cpu:*"
    if device_option.device_type == caffe2_pb2.CUDA:
        return "/gpu:{}".format(device_option.device_id)
    raise Exception("Unhandled device", device_option) 

def _tf_device(device_option):
    '''
    Handle the devices.

    Args:
        device_option (caffe2_pb2.DeviceOption): DeviceOption protobuf,
            associated to an operator, that contains information such as
            device_type (optional), cuda_gpu_id (optional), node_name (optional,
            tells which node the operator should execute on). See caffe2.proto
            in caffe2/proto for the full list.

    Returns:
        Formatted string representing device information contained in
            device_option.
    '''
    if not device_option.HasField("device_type"):
        return ""
    if device_option.device_type == caffe2_pb2.CPU or device_option.device_type == caffe2_pb2.MKLDNN:
        return "/cpu:*"
    if device_option.device_type == caffe2_pb2.CUDA:
        return "/gpu:{}".format(device_option.device_id)
    raise Exception("Unhandled device", device_option) 

def __init__(self, device_option: DeviceOption):
        super(Caffe2Network, self).__init__()
        self.device_option = device_option

        self.train_model = model_helper.ModelHelper(name="train_default_net")
        self.test_model = model_helper.ModelHelper(name="test_default_net", init_params=False)
        self.train_net = self.train_model.net
        self.test_net = self.test_model.net
        self.train_init_net = self.train_model.param_init_net
        self.test_init_net = self.test_model.param_init_net
        self.workspace = workspace
        self.output_dict = {}
        self.param_names = None
        # dict that helps us remember that we already added the gradients to the graph for a given loss
        self.gradients_by_loss = {}
        self.is_cuda = (device_option.device_type == caffe2_pb2.CUDA) 

def create_const_fill_op(
    name: str,
    blob: Union[np.ndarray, workspace.Int8Tensor],
    device_option: Optional[caffe2_pb2.DeviceOption] = None,
) -> caffe2_pb2.OperatorDef:
    """
    Given a blob object, return the Caffe2 operator that creates this blob
    as constant. Currently support NumPy tensor and Caffe2 Int8Tensor.
    """

    tensor_type = type(blob)
    assert tensor_type in [
        np.ndarray,
        workspace.Int8Tensor,
    ], 'Error when creating const fill op for "{}", unsupported blob type: {}'.format(
        name, type(blob)
    )

    if tensor_type == np.ndarray:
        return _create_const_fill_op_from_numpy(name, blob, device_option)
    elif tensor_type == workspace.Int8Tensor:
        assert device_option is None
        return _create_const_fill_op_from_c2_int8_tensor(name, blob) 

def construct_init_net_from_params(
    params: Dict[str, Any], device_options: Optional[Dict[str, caffe2_pb2.DeviceOption]] = None
) -> caffe2_pb2.NetDef:
    """
    Construct the init_net from params dictionary
    """
    init_net = caffe2_pb2.NetDef()
    device_options = device_options or {}
    for name, blob in params.items():
        if isinstance(blob, str):
            logger.warning(
                (
                    "Blob {} with type {} is not supported in generating init net,"
                    " skipped.".format(name, type(blob))
                )
            )
            continue
        init_net.op.extend(
            [create_const_fill_op(name, blob, device_option=device_options.get(name, None))]
        )
        init_net.external_output.append(name)
    return init_net 

def get_params_from_init_net(
    init_net: caffe2_pb2.NetDef,
) -> [Dict[str, Any], Dict[str, caffe2_pb2.DeviceOption]]:
    """
    Take the output blobs from init_net by running it.
    Outputs:
        params: dict from blob name to numpy array
        device_options: dict from blob name to the device option of its creating op
    """
    # NOTE: this assumes that the params is determined by producer op with the
    # only exception be CopyGPUToCPU which is CUDA op but returns CPU tensor.
    def _get_device_option(producer_op):
        if producer_op.type == "CopyGPUToCPU":
            return caffe2_pb2.DeviceOption()
        else:
            return producer_op.device_option

    with ScopedWS("__get_params_from_init_net__", is_reset=True, is_cleanup=True) as ws:
        ws.RunNetOnce(init_net)
        params = {b: fetch_any_blob(b) for b in init_net.external_output}
    ssa, versions = core.get_ssa(init_net)
    producer_map = get_producer_map(ssa)
    device_options = {
        b: _get_device_option(init_net.op[producer_map[(b, versions[b])][0]])
        for b in init_net.external_output
    }
    return params, device_options 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.cuda_gpu_id = gpu_id
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.cuda_gpu_id = gpu_id
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.cuda_gpu_id = gpu_id
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.device_id = gpu_id
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.device_id = gpu_id
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.cuda_gpu_id = gpu_id
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.cuda_gpu_id = gpu_id
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def __init__(self, model: d5.ops.OnnxModel, device_option: DeviceOption,
                 events: List[d5.ExecutorEvent] = []):
        super(Caffe2GraphExecutor, self).__init__(Caffe2Network(device_option), events)
        self.device_option = device_option
        self.model = model

        with core.DeviceScope(self.device_option):
            model.accept(Caffe2Visitor(device_option), self.network) 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.cuda_gpu_id = gpu_id
    return device_option 

def get_device_option_cpu():
    device_option = core.DeviceOption(caffe2_pb2.CPU)
    return device_option 

def Caffe2ToOnnx(init_def, predict_def, data_shape):
    """transfer caffe2 to onnx"""
    from caffe2.proto import caffe2_pb2
    from caffe2.python.onnx import frontend
    from caffe2.python import workspace

    old_ws_name = workspace.CurrentWorkspace()
    workspace.SwitchWorkspace("_onnx_porting_", True)

    data_type = onnx.TensorProto.FLOAT
    value_info = {
        str(predict_def.op[0].input[0]) : (data_type, data_shape)
    }

    device_opts_cpu = caffe2_pb2.DeviceOption()
    device_opts_cpu.device_type = caffe2_pb2.CPU
    UpdateDeviceOption(device_opts_cpu, init_def)
    UpdateDeviceOption(device_opts_cpu, predict_def)

    onnx_model = frontend.caffe2_net_to_onnx_model(
        predict_def,
        init_def,
        value_info
    )

    onnx.checker.check_model(onnx_model)
    workspace.SwitchWorkspace(old_ws_name)
    return onnx_model 

def get_device_option_cuda(gpu_id=0):
    device_option = caffe2_pb2.DeviceOption()
    device_option.device_type = caffe2_pb2.CUDA
    device_option.device_id = gpu_id
    return device_option 

def convert_net(args, net, blobs):

    @op_filter()
    def convert_op_name(op):
        if args.device != 'gpu':
            if op.engine != 'DEPTHWISE_3x3':
                op.engine = ''
            op.device_option.CopyFrom(caffe2_pb2.DeviceOption())
        reset_names(op.input)
        reset_names(op.output)
        return [op]

    @op_filter(type="Python", inputs=['rpn_cls_probs', 'rpn_bbox_pred', 'im_info'])
    def convert_gen_proposal(op_in):
        gen_proposals_op, ext_input = convert_gen_proposals(
            op_in, blobs,
            rpn_min_size=float(cfg.TEST.RPN_MIN_SIZE),
            rpn_post_nms_topN=cfg.TEST.RPN_POST_NMS_TOP_N,
            rpn_pre_nms_topN=cfg.TEST.RPN_PRE_NMS_TOP_N,
            rpn_nms_thres=cfg.TEST.RPN_NMS_THRESH,
        )
        net.external_input.extend([ext_input])
        return [gen_proposals_op]

    @op_filter(input_has='rois')
    def convert_rpn_rois(op):
        for j in range(0, len(op.input)):
            if op.input[j] == 'rois':
                print('Converting op {} input name: rois -> rpn_rois:\n{}'.format(
                    op.type, op))
                op.input[j] = 'rpn_rois'
        return [op]

    @op_filter(type_in=['StopGradient', 'Alias'])
    def convert_remove_op(op):
        print('Removing op {}:\n{}'.format(op.type, op))
        return []

    convert_op_in_proto(net, convert_op_name)
    convert_op_in_proto(net, [
        convert_gen_proposal, convert_rpn_rois, convert_remove_op
    ])

    reset_names(net.external_input)
    reset_names(net.external_output)

    reset_blob_names(blobs) 

def convert_net(args, net, blobs):

    @op_filter()
    def convert_op_name(op):
        if args.device != 'gpu':
            if op.engine != 'DEPTHWISE_3x3':
                op.engine = ''
            op.device_option.CopyFrom(caffe2_pb2.DeviceOption())
        reset_names(op.input)
        reset_names(op.output)
        return [op]

    @op_filter(type="Python", inputs=['rpn_cls_probs', 'rpn_bbox_pred', 'im_info'])
    def convert_gen_proposal(op_in):
        gen_proposals_op, ext_input = convert_gen_proposals(
            op_in, blobs,
            rpn_min_size=float(cfg.TEST.RPN_MIN_SIZE),
            rpn_post_nms_topN=cfg.TEST.RPN_POST_NMS_TOP_N,
            rpn_pre_nms_topN=cfg.TEST.RPN_PRE_NMS_TOP_N,
            rpn_nms_thres=cfg.TEST.RPN_NMS_THRESH,
        )
        net.external_input.extend([ext_input])
        return [gen_proposals_op]

    @op_filter(input_has='rois')
    def convert_rpn_rois(op):
        for j in range(0, len(op.input)):
            if op.input[j] == 'rois':
                print('Converting op {} input name: rois -> rpn_rois:\n{}'.format(
                    op.type, op))
                op.input[j] = 'rpn_rois'
        return [op]

    @op_filter(type_in=['StopGradient', 'Alias'])
    def convert_remove_op(op):
        print('Removing op {}:\n{}'.format(op.type, op))
        return []

    convert_op_in_proto(net, convert_op_name)
    convert_op_in_proto(net, [
        convert_gen_proposal, convert_rpn_rois, convert_remove_op
    ])

    reset_names(net.external_input)
    reset_names(net.external_output)

    reset_blob_names(blobs)