[Python API] Support of FP16 blobs (openvinotoolkit#3893)

* [Python API] Support of FP16 blobs

* test_Blob refactoring

* support fp16 for exec_net.infer method

* add precisions

Co-authored-by: anastasia.kuporosova <[email protected]>

inference-engine/ie_bridges/python/src/openvino/inference_engine/constants.pyx

@@ -18,7 +18,7 @@ from .cimport ie_api_impl_defs as C
 import numpy as np
 from enum import Enum
 
-supported_precisions = ["FP32", "FP64", "FP16", "I64", "U64", "I32", "U32", "I16", "I8", "U16", "U8"]
+supported_precisions = ["FP32", "FP64", "FP16", "I64", "U64", "I32", "U32", "I16", "I8", "U16", "U8", "BOOL"]
 
 known_plugins = ['CPU', 'GPU', 'FPGA', 'MYRIAD', 'HETERO', 'HDDL', 'MULTI']
 
@@ -34,7 +34,8 @@ format_map = {
       'U16'  : np.uint16,
       'I8'   : np.int8,
       'U8'   : np.uint8,
-      'I64'  : np.int64
+      'I64'  : np.int64,
+      'BOOL' : np.uint8
     }
 
 layout_str_to_enum = {'ANY': C.Layout.ANY,

inference-engine/ie_bridges/python/src/openvino/inference_engine/ie_api.pyx

@@ -173,7 +173,8 @@ cdef class Blob:
                 fp64_array_memview = self._array_data
                 self._ptr = C.make_shared_blob[double](c_tensor_desc, &fp64_array_memview[0], fp64_array_memview.shape[0])
             elif precision == "FP16":
-                raise RuntimeError("Currently, it's impossible to set_blob with FP16 precision")
+                I16_array_memview = self._array_data.view(dtype=np.int16)
+                self._ptr = C.make_shared_blob[int16_t](c_tensor_desc, &I16_array_memview[0], I16_array_memview.shape[0])
             elif precision == "I16":
                 I16_array_memview = self._array_data
                 self._ptr = C.make_shared_blob[int16_t](c_tensor_desc, &I16_array_memview[0], I16_array_memview.shape[0])
@@ -1222,7 +1223,10 @@ cdef class InferRequest:
     def _fill_inputs(self, inputs):
         for k, v in inputs.items():
             assert k in self._inputs_list, f"No input with name {k} found in network"
-            self.input_blobs[k].buffer[:] = v
+            if self.input_blobs[k].tensor_desc.precision == "FP16":
+                self.input_blobs[k].buffer[:] = v.view(dtype=np.int16)
+            else:
+                self.input_blobs[k].buffer[:] = v
 
 
 ## This class contains the information about the network model read from IR and allows you to manipulate with

inference-engine/ie_bridges/python/tests/test_Blob.py

@@ -36,87 +36,48 @@ def test_get_buffer():
     blob = Blob(tensor_desc, array)
     assert np.array_equal(blob.buffer, array)
 
-
-def test_write_to_buffer_fp32():
-    tensor_desc = TensorDesc("FP32", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.float32)
+def write_to_buffer(precision, numpy_precision):
+    tensor_desc = TensorDesc(precision, [1, 3, 127, 127], "NCHW")
+    array = np.zeros(shape=(1, 3, 127, 127), dtype=numpy_precision)
     blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.float32)
+    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=numpy_precision)
     blob.buffer[:] = ones_arr
     assert np.array_equal(blob.buffer, ones_arr)
 
+def test_write_to_buffer_fp32():
+    write_to_buffer("FP32", np.float32)
+
 
 def test_write_to_buffer_fp64():
-    tensor_desc = TensorDesc("FP64", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.float64)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.float64)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
+    write_to_buffer("FP64", np.float64)
 
 
-@pytest.mark.skip(reason="Need to figure out how to implement right conversion")
 def test_write_to_buffer_fp16():
-    tensor_desc = TensorDesc("FP16", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.float16)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.float16)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
+    write_to_buffer("FP16", np.float16)
 
 
 def test_write_to_buffer_int8():
-    tensor_desc = TensorDesc("I8", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.int8)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.int8)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
+    write_to_buffer("I8", np.int8)
 
 
 def test_write_to_buffer_uint8():
-    tensor_desc = TensorDesc("U8", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.uint8)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.uint8)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
+    write_to_buffer("U8", np.uint8)
 
 
 def test_write_to_buffer_int32():
-    tensor_desc = TensorDesc("I32", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.int32)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.int32)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
+    write_to_buffer("I32", np.int32)
 
 
 def test_write_to_buffer_int16():
-    tensor_desc = TensorDesc("I16", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.int16)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.int16)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
+    write_to_buffer("I16", np.int16)
 
 
 def test_write_to_buffer_uint16():
-    tensor_desc = TensorDesc("U16", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.uint16)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.uint16)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
+    write_to_buffer("U16", np.uint16)
 
 
 def test_write_to_buffer_int64():
-    tensor_desc = TensorDesc("I64", [1, 3, 127, 127], "NCHW")
-    array = np.zeros(shape=(1, 3, 127, 127), dtype=np.int64)
-    blob = Blob(tensor_desc, array)
-    ones_arr = np.ones(shape=(1, 3, 127, 127), dtype=np.int64)
-    blob.buffer[:] = ones_arr
-    assert np.array_equal(blob.buffer, ones_arr)
+    write_to_buffer("I64", np.int64)
 
 
 def test_write_numpy_scalar_int64():