def get_output_fmt(self):
output_info = ffi.new('bmnet_output_info_t *')
lib.bmnet_get_output_info(self.net[0], output_info)
output_fmt = []
for i in range(output_info.output_num):
output_fmt.append(output_info.fmt_array[i])
return output_fmt
def get_output_threshold(self):
output_info = ffi.new('bmnet_output_info_t *')
lib.bmnet_get_output_info(self.net[0], output_info)
output_threshold = []
for i in range(output_info.output_num):
output_threshold.append(output_info.threshold_array[i])
return output_threshold
def set_input_shape(self, input_shape):
shape = ffi.new('shape_t *')
shape.dim = input_shape.shape[0]
shape.n = input_shape[0]
shape.c = input_shape[1]
shape.h = input_shape[2]
shape.w = input_shape[3]
lib.bmnet_set_input_shape(self.net[0], shape[0])
def get_output_shape(self):
output_info = ffi.new('bmnet_output_info_t *')
lib.bmnet_get_output_info(self.net[0], output_info)
output_shape = []
for i in range(output_info.output_num):
shape = output_info.shape_array[i]
if shape.dim == 1:
output_shape.append([shape.n])
elif shape.dim == 2:
output_shape.append([shape.n, shape.c])
elif shape.dim == 3:
output_shape.append([shape.n, shape.c, shape.h])
elif shape.dim == 4:
output_shape.append([shape.n, shape.c, shape.h, shape.w])
return output_shape
def set_input_shapes(self, input_shapes):
shape_arr = []
for i in range(len(input_shapes)):
shape = ffi.new('shape_t *')
shape.dim = input_shapes[i].shape[0]
if shape.dim == 1:
shape.n = input_shapes[i][0]
elif shape.dim == 2:
shape.n = input_shapes[i][0]
shape.c = input_shapes[i][1]
elif shape.dim == 3:
shape.n = input_shapes[i][0]
shape.c = input_shapes[i][1]
shape.h = input_shapes[i][2]
elif shape.dim == 4:
shape.n = input_shapes[i][0]
shape.c = input_shapes[i][1]
shape.h = input_shapes[i][2]
shape.w = input_shapes[i][3]
shape_arr.append(shape[0])
lib.bmnet_set_input_shape2(self.net[0], shape_arr, len(input_shapes))
def __init__(self, index):
self.ctx = ffi.new('bmctx_t *')
lib.bm_init(index, self.ctx)
def get_output_size(self):
output_info = ffi.new('bmnet_output_info_t *')
lib.bmnet_get_output_info(self.net[0], output_info)
return output_info.output_size
def inference(self, input, output):
output_info = ffi.new('bmnet_output_info_t *')
lib.bmnet_get_output_info(self.net[0], output_info)
lib.bmnet_inference(self.net[0], ffi.from_buffer(input),
ffi.from_buffer(output))
def __init__(self, bmctx, bmodel_path):
self.net = ffi.new('bmnet_t *')
bmodel = ffi.new('char[]', bmodel_path)
lib.bmnet_register_bmodel(bmctx.ctx[0], bmodel, self.net)