Added Additional Operators to support for Requantizer; Resolved assertion failed due to unequal shapes

tensorflow/lite/micro/tools/model_transforms_utils.py

@@ -246,7 +246,7 @@ def clear_resource_variable_buffers(model):
     buffer_idx = tensor.buffer
     if (tensor.type != schema_fb.TensorType.RESOURCE
         and buffer_idx not in multi_subgraph_resource_buffers
-        and model.buffers[buffer_idx].data != []):
+        and np.array(model.buffers[buffer_idx].data).size != 0):
       # if the entire initialization subgraph has not been cleared, we cannot
       # make any additional changes to the flatbuffer
       return

tensorflow/lite/micro/tools/requantize_flatbuffer.py

@@ -71,25 +71,51 @@
     requantize_flatbuffer_utils.requantize_fully_connected,
     schema_py_generated.BuiltinOperator.TRANSPOSE_CONV:
     requantize_flatbuffer_utils.requantize_transpose_conv,
+    schema_py_generated.BuiltinOperator.GREATER:
+    requantize_flatbuffer_utils.requantize_greater,
+    schema_py_generated.BuiltinOperator.SUB:
+    requantize_flatbuffer_utils.requantize_sub,
+    schema_py_generated.BuiltinOperator.MUL:
+    requantize_flatbuffer_utils.requantize_mul,
+    schema_py_generated.BuiltinOperator.BATCH_MATMUL:
+    requantize_flatbuffer_utils.requantize_batch_matmul,
+    schema_py_generated.BuiltinOperator.SELECT_V2:
+    requantize_flatbuffer_utils.requantize_select_v2,
+    schema_py_generated.BuiltinOperator.CONCATENATION:
+    requantize_flatbuffer_utils.requantize_concatenation,
+    schema_py_generated.BuiltinOperator.ADD:
+    requantize_flatbuffer_utils.requantize_add,
+    schema_py_generated.BuiltinOperator.PAD:
+    requantize_flatbuffer_utils.requantize_pad,
 }
 
 # List of tested simple operators (no weight and bias, e.g., reshape) see tensorflow/lite/schema/schema.fbs for op code names
 _TESTED_SIMPLE_OPS = [
-    schema_py_generated.BuiltinOperator.ADD,
     schema_py_generated.BuiltinOperator.CONCATENATION,
     schema_py_generated.BuiltinOperator.DEQUANTIZE,
     schema_py_generated.BuiltinOperator.LEAKY_RELU,
     schema_py_generated.BuiltinOperator.LOGISTIC,
     schema_py_generated.BuiltinOperator.MEAN,
-    schema_py_generated.BuiltinOperator.MUL,
-    schema_py_generated.BuiltinOperator.PAD,
     schema_py_generated.BuiltinOperator.QUANTIZE,
     schema_py_generated.BuiltinOperator.RESHAPE,
     schema_py_generated.BuiltinOperator.RSQRT,
     schema_py_generated.BuiltinOperator.SQRT,
     schema_py_generated.BuiltinOperator.SQUARED_DIFFERENCE,
     schema_py_generated.BuiltinOperator.STRIDED_SLICE,
     schema_py_generated.BuiltinOperator.SUB,
+    schema_py_generated.BuiltinOperator.CALL_ONCE,
+    schema_py_generated.BuiltinOperator.VAR_HANDLE,
+    schema_py_generated.BuiltinOperator.READ_VARIABLE,
+    schema_py_generated.BuiltinOperator.ASSIGN_VARIABLE,
+    schema_py_generated.BuiltinOperator.FLOOR_DIV,
+    schema_py_generated.BuiltinOperator.CAST,
+    schema_py_generated.BuiltinOperator.COS,
+    schema_py_generated.BuiltinOperator.SIN,
+    schema_py_generated.BuiltinOperator.UNPACK,
+    schema_py_generated.BuiltinOperator.TRANSPOSE,  
+    schema_py_generated.BuiltinOperator.SPLIT,
+    schema_py_generated.BuiltinOperator.RESIZE_NEAREST_NEIGHBOR,
+    schema_py_generated.BuiltinOperator.SELECT_V2,
 ]
 
 _SUPPORTED_OPS = set(

tensorflow/lite/micro/tools/requantize_flatbuffer_utils.py

@@ -265,6 +265,101 @@ def requantize_fully_connected(tensors, buffers, op):
     set_bias_type_int64(buffers, input_tensor, weight_tensor, bias_tensor)
 
 
+def requantize_greater(tensors, buffers, op):
+  # Indices are from tensorflow/lite/micro/kernels/fully_connected_common.cc
+  input1_tensor = tensors[op.inputs[0]]
+  # weight stays the same, no change needed
+  input2_tensor = tensors[op.inputs[1]]
+  output_tensor = tensors[op.outputs[0]]
+
+  change_activation_tensor_8to16(input1_tensor, buffers)
+  change_activation_tensor_8to16(input2_tensor, buffers)
+  change_activation_tensor_8to16(output_tensor, buffers)
+
+
+def requantize_select_v2(tensors, buffers, op):
+  # Indices are from tensorflow/lite/micro/kernels/fully_connected_common.cc
+  input1_tensor = tensors[op.inputs[0]]
+  # weight stays the same, no change needed
+  input2_tensor = tensors[op.inputs[1]]
+  input3_tensor = tensors[op.inputs[2]]
+  output_tensor = tensors[op.outputs[0]]
+
+  change_activation_tensor_8to16(input1_tensor, buffers)
+  change_activation_tensor_8to16(input2_tensor, buffers)
+  change_activation_tensor_8to16(input3_tensor, buffers)
+  change_activation_tensor_8to16(output_tensor, buffers)
+
+
+def requantize_pad(tensors, buffers, op):
+  # Indices are from tensorflow/lite/micro/kernels/fully_connected_common.cc
+  input1_tensor = tensors[op.inputs[0]]
+  # weight stays the same, no change needed
+  input2_tensor = tensors[op.inputs[1]]
+  # input3_tensor = tensors[op.inputs[2]]
+  output_tensor = tensors[op.outputs[0]]
+
+  change_activation_tensor_8to16(input1_tensor, buffers)
+  change_activation_tensor_8to16(input2_tensor, buffers)
+  # change_activation_tensor_8to16(input3_tensor, buffers)
+  change_activation_tensor_8to16(output_tensor, buffers)
+
+
+def requantize_sub(tensors, buffers, op):
+  # Indices are from tensorflow/lite/micro/kernels/sub.cc
+  input1_tensor = tensors[op.inputs[0]]
+  input2_tensor = tensors[op.inputs[1]]
+  output_tensor = tensors[op.outputs[0]]
+
+  change_activation_tensor_8to16(input1_tensor, buffers)
+  change_activation_tensor_8to16(input2_tensor, buffers)
+  change_activation_tensor_8to16(output_tensor, buffers)
+
+
+def requantize_batch_matmul(tensors, buffers, op):
+  # Indices are from tensorflow/lite/micro/kernels/sub.cc
+  input1_tensor = tensors[op.inputs[0]]
+  input2_tensor = tensors[op.inputs[1]]
+  output_tensor = tensors[op.outputs[0]]
+
+  change_activation_tensor_8to16(input1_tensor, buffers)
+  change_activation_tensor_8to16(input2_tensor, buffers)
+  change_activation_tensor_8to16(output_tensor, buffers)
+
+
+def requantize_concatenation(tensors, buffers, op):
+  # Indices are from tensorflow/lite/micro/kernels/sub.cc
+  input1_tensor = tensors[op.inputs[0]]
+  input2_tensor = tensors[op.inputs[1]]
+  output_tensor = tensors[op.outputs[0]]
+
+  change_activation_tensor_8to16(input1_tensor, buffers)
+  change_activation_tensor_8to16(input2_tensor, buffers)
+  change_activation_tensor_8to16(output_tensor, buffers)
+
+
+def requantize_mul(tensors, buffers, op):
+  # Indices are from tensorflow/lite/micro/kernels/sub.cc
+  input1_tensor = tensors[op.inputs[0]]
+  input2_tensor = tensors[op.inputs[1]]
+  output_tensor = tensors[op.outputs[0]]
+
+  change_activation_tensor_8to16(input1_tensor, buffers)
+  change_activation_tensor_8to16(input2_tensor, buffers)
+  change_activation_tensor_8to16(output_tensor, buffers)
+
+
+def requantize_add(tensors, buffers, op):
+  # Indices are from tensorflow/lite/micro/kernels/sub.cc
+  input1_tensor = tensors[op.inputs[0]]
+  input2_tensor = tensors[op.inputs[1]]
+  output_tensor = tensors[op.outputs[0]]
+
+  change_activation_tensor_8to16(input1_tensor, buffers)
+  change_activation_tensor_8to16(input2_tensor, buffers)
+  change_activation_tensor_8to16(output_tensor, buffers)
+
+
 def requantize_unidirectional_sequence_lstm(tensors, buffers, op):
   """Requantize the unidirectonal sequance lstm op from int8 to int16 """
   input_tensor = tensors[op.inputs[0]]