onnx2tf 1.28.6__py3-none-any.whl → 1.28.8__py3-none-any.whl

onnx2tf/__init__.py

@@ -1,3 +1,3 @@
 from onnx2tf.onnx2tf import convert, main
 
-__version__ = '1.28.6'
+__version__ = '1.28.8'

onnx2tf/onnx2tf.py

@@ -74,6 +74,8 @@ def convert(
     copy_onnx_input_output_names_to_tflite: Optional[bool] = False,
     output_dynamic_range_quantized_tflite: Optional[bool] = False,
     output_integer_quantized_tflite: Optional[bool] = False,
+    quant_norm_mean: Optional[str] = '[[[[0.485, 0.456, 0.406]]]]',
+    quant_norm_std: Optional[str] = '[[[[0.229, 0.224, 0.225]]]]',
     quant_type: Optional[str] = 'per-channel',
     custom_input_op_name_np_data_path: Optional[List] = None,
     input_quant_dtype: Optional[str] = 'int8',
@@ -172,6 +174,16 @@ def convert(
     output_integer_quantized_tflite: Optional[bool]
         Output of integer quantized tflite.
 
+    quant_norm_mean: Optional[str]
+        Normalized average value during quantization.\n
+        Only valid when the "-cind" option is not used.\n
+        Default: "[[[[0.485, 0.456, 0.406]]]]"
+
+    quant_norm_std: Optional[str]
+        Normalized standard deviation during quantization.\n
+        Only valid when the "-cind" option is not used.\n
+        Default: "[[[[0.229, 0.224, 0.225]]]]"
+
     quant_type: Optional[str]
         Selects whether "per-channel" or "per-tensor" quantization is used.\n
         Default: "per-channel"
@@ -621,6 +633,19 @@ def convert(
         )
         sys.exit(1)
 
+    # Normalized average value during quantization
+    if quant_norm_mean:
+        quant_norm_mean_np = np.array(ast.literal_eval(quant_norm_mean), dtype=np.float32)
+    else:
+        quant_norm_mean_np = np.array(ast.literal_eval("[[[[0.000, 0.000, 0.000]]]]"), dtype=np.float32)
+
+    # Normalized standard deviation during quantization
+    if quant_norm_std:
+        quant_norm_std_np = np.array(ast.literal_eval(quant_norm_std), dtype=np.float32)
+    else:
+        quant_norm_std_np = np.array(ast.literal_eval("[[[[1.000, 1.000, 1.000]]]]"), dtype=np.float32)
+
+    # param replacement
     replacement_parameters = None
     if param_replacement_file:
         if not os.path.isfile(param_replacement_file):
@@ -1683,6 +1708,8 @@ def convert(
                     Color.BLUE(f'shape') + f': {output_shape} '+
                     Color.BLUE(f'dtype') + f': {output_dtype}'
                 )
+            info(Color.BLUE(f'quant_norm_mean') + f': {quant_norm_mean_np} ')
+            info(Color.BLUE(f'quant_norm_std') + f': {quant_norm_std_np} ')
             print('')
 
             # INT8 Converter
@@ -1720,18 +1747,20 @@ def convert(
 
                     if model_input.shape[-1] == 3:
                         # RGB
-                        mean = np.asarray([[[[0.485, 0.456, 0.406]]]], dtype=np.float32)
-                        std = np.asarray([[[[0.229, 0.224, 0.225]]]], dtype=np.float32)
+                        mean = quant_norm_mean_np
+                        std = quant_norm_std_np
                     elif model_input.shape[-1] == 4:
                         # RGBA
-                        mean = np.asarray([[[[0.485, 0.456, 0.406, 0.000]]]], dtype=np.float32)
-                        std = np.asarray([[[[0.229, 0.224, 0.225, 1.000]]]], dtype=np.float32)
+                        zero = np.zeros((*quant_norm_mean_np.shape[:-1], 1), dtype=quant_norm_mean_np.dtype)
+                        mean = np.concatenate([quant_norm_mean_np, zero], axis=-1)
+                        one = np.ones((*quant_norm_std_np.shape[:-1], 1), dtype=quant_norm_std_np.dtype)
+                        std = np.concatenate([quant_norm_std_np, zero], axis=-1)
                         new_element_array = np.full((*calib_data.shape[:-1], 1), 0.500, dtype=np.float32)
                         calib_data = np.concatenate((calib_data, new_element_array), axis=-1)
                     else:
                         # Others
-                        mean = np.asarray([[[[0.485, 0.456, 0.406]]]], dtype=np.float32)
-                        std = np.asarray([[[[0.229, 0.224, 0.225]]]], dtype=np.float32)
+                        mean = quant_norm_mean_np
+                        std = quant_norm_std_np
 
                     calib_data_dict[model_input.name] = \
                         [
@@ -2475,6 +2504,24 @@ def main():
             'Selects whether "per-channel" or "per-tensor" quantization is used. \n' +
             'Default: "per-channel"'
     )
+    parser.add_argument(
+        '-qnm',
+        '--quant_norm_mean',
+        type=str,
+        default='[[[[0.485, 0.456, 0.406]]]]',
+        help=\
+            'Normalized average value during quantization. \n' +
+            'Default: "[[[[0.485, 0.456, 0.406]]]]"'
+    )
+    parser.add_argument(
+        '-qns',
+        '--quant_norm_std',
+        type=str,
+        default='[[[[0.229, 0.224, 0.225]]]]',
+        help=\
+            'Normalized standard deviation during quantization. \n' +
+            'Default: "[[[[0.229, 0.224, 0.225]]]]"'
+    )
     parser.add_argument(
         '-cind',
         '--custom_input_op_name_np_data_path',
@@ -3061,6 +3108,8 @@ def main():
         copy_onnx_input_output_names_to_tflite=args.copy_onnx_input_output_names_to_tflite,
         output_dynamic_range_quantized_tflite=args.output_dynamic_range_quantized_tflite,
         output_integer_quantized_tflite=args.output_integer_quantized_tflite,
+        quant_norm_mean=args.quant_norm_mean,
+        quant_norm_std=args.quant_norm_std,
         quant_type=args.quant_type,
         custom_input_op_name_np_data_path=custom_params,
         input_quant_dtype=args.input_quant_dtype,

onnx2tf/ops/Conv.py

@@ -62,7 +62,7 @@ def make_node(
         before_op_output_shape_trans,
     )
     kernel_shape = graph_node.attrs.get('kernel_shape', [])
-    kernel_size = len(kernel_shape)
+    kernel_size = len(kernel_shape) if kernel_shape != [] else len(graph_node.inputs[1].shape) - 2
     try:
         input_weights = get_weights_constant_or_variable(
             const_or_var=graph_node.inputs[1] \

{onnx2tf-1.28.6.dist-info → onnx2tf-1.28.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: onnx2tf
-Version: 1.28.6
+Version: 1.28.8
 Summary: Self-Created Tools to convert ONNX files (NCHW) to TensorFlow/TFLite/Keras format (NHWC). The purpose of this tool is to solve the massive Transpose extrapolation problem in onnx-tensorflow (onnx-tf).
 Home-page: https://github.com/PINTO0309/onnx2tf
 Author: Katsuya Hyodo
@@ -334,7 +334,7 @@ Video speed is adjusted approximately 50 times slower than actual speed.
   docker run --rm -it \
   -v `pwd`:/workdir \
   -w /workdir \
-  ghcr.io/pinto0309/onnx2tf:1.28.6
+  ghcr.io/pinto0309/onnx2tf:1.28.8
 
   or
 
@@ -342,7 +342,7 @@ Video speed is adjusted approximately 50 times slower than actual speed.
   docker run --rm -it \
   -v `pwd`:/workdir \
   -w /workdir \
-  docker.io/pinto0309/onnx2tf:1.28.6
+  docker.io/pinto0309/onnx2tf:1.28.8
 
   or
 
@@ -1667,6 +1667,16 @@ optional arguments:
     Selects whether "per-channel" or "per-tensor" quantization is used.
     Default: "per-channel"
 
+  -qnm QUANT_NORM_MEAN, --quant_norm_mean QUANT_NORM_MEAN
+    Normalized average value during quantization.
+    Only valid when the "-cind" option is not used.
+    Default: "[[[[0.485, 0.456, 0.406]]]]"
+
+  -qns QUANT_NORM_STD, --quant_norm_std QUANT_NORM_STD
+    Normalized standard deviation during quantization.
+    Only valid when the "-cind" option is not used.
+    Default: "[[[[0.229, 0.224, 0.225]]]]"
+
   -cind INPUT_NAME NUMPY_FILE_PATH MEAN STD, \
     --custom_input_op_name_np_data_path INPUT_NAME NUMPY_FILE_PATH MEAN STD
     Input name of OP and path of data file (Numpy) for custom input for -cotof or -oiqt,
@@ -2095,6 +2105,8 @@ convert(
   output_weights: Optional[bool] = False,
   copy_onnx_input_output_names_to_tflite: Optional[bool] = False,
   output_integer_quantized_tflite: Optional[bool] = False,
+  quant_norm_mean: Optional[str] = '[[[[0.485, 0.456, 0.406]]]]',
+  quant_norm_std: Optional[str] = '[[[[0.229, 0.224, 0.225]]]]',
   quant_type: Optional[str] = 'per-channel',
   custom_input_op_name_np_data_path: Optional[List] = None,
   input_quant_dtype: Optional[str] = 'int8',
@@ -2190,6 +2202,16 @@ convert(
     output_integer_quantized_tflite: Optional[bool]
       Output of integer quantized tflite.
 
+    quant_norm_mean: Optional[str]
+        Normalized average value during quantization.
+        Only valid when the "-cind" option is not used.
+        Default: "[[[[0.485, 0.456, 0.406]]]]"
+
+    quant_norm_std: Optional[str]
+        Normalized standard deviation during quantization.
+        Only valid when the "-cind" option is not used.
+        Default: "[[[[0.229, 0.224, 0.225]]]]"
+
     quant_type: Optional[str]
       Selects whether "per-channel" or "per-tensor" quantization is used.
       Default: "per-channel"

{onnx2tf-1.28.6.dist-info → onnx2tf-1.28.8.dist-info}/RECORD

@@ -1,6 +1,6 @@
-onnx2tf/__init__.py,sha256=GyshUypwrMRTRG2JuL0N2SN72_h_yEdaXSrMZjlRtm8,66
+onnx2tf/__init__.py,sha256=PpTfjKEZ7PWIDHIvtSVU9Epc3rkkmE02EOIWvlHFnWU,66
 onnx2tf/__main__.py,sha256=2RSCQ7d4lc6CwD-rlGn9UicPFg-P5du7ZD_yh-kuBEU,57
-onnx2tf/onnx2tf.py,sha256=iQwYuLj1f7gHWU9r_4L1pyRnMsJgHkqEPeavxejtY4Y,146964
+onnx2tf/onnx2tf.py,sha256=v6dx8ZQrj2h4-xv3lae4wkdts1NO3tEINS_sK0dykvw,148911
 onnx2tf/ops/Abs.py,sha256=V7btmCG_ZvK_qJovUsguq0ZMJ349mhNQ4FHSgzP_Yuo,4029
 onnx2tf/ops/Acos.py,sha256=Fo8YkFKuWq8Fi2xUrBdKcAH1yJ8r5pjSD0wgLttTNdk,4003
 onnx2tf/ops/Acosh.py,sha256=ATQj2cT5JS_mTfXi0kXqJ1yzSZu5J0zHA5VjV3j7uKY,3588
@@ -26,7 +26,7 @@ onnx2tf/ops/Concat.py,sha256=CKfJbiAwP7h9sFFVyueHJCbwMkUo3NXqkTuRc8v7Tw8,31215
 onnx2tf/ops/ConcatFromSequence.py,sha256=z8pNmGQRGq9cxWORW330NZS_0zsmhFudLswMyPn8AXU,3086
 onnx2tf/ops/Constant.py,sha256=BNZLzNI4rK9kXgVWwD-2RFsDsH7mMy7AY2JSgTNXIWk,10696
 onnx2tf/ops/ConstantOfShape.py,sha256=6eYm-niow-6fHVEyNyi81BdrVe3IbcdazCp2nySWExA,2331
-onnx2tf/ops/Conv.py,sha256=jcBIgZY4IXxVnlNNmkGhIVbsE6UWmA7xfwe9Jw3J1G8,37881
+onnx2tf/ops/Conv.py,sha256=MX4bDOZUnGpLGQkqVRunc_iw7TU-Rb4pgX1rE64bR1k,37944
 onnx2tf/ops/ConvInteger.py,sha256=UVHy1de6uCLM7IXP4vpzPz2pN2ej278lGBiqTPNdXFA,26664
 onnx2tf/ops/ConvTranspose.py,sha256=C7CR6m3kz0MtUBdtWrrKWZbZL7tJpGXl7Nkn3DRiEaA,15410
 onnx2tf/ops/Cos.py,sha256=0v5ZJZRzrswVEObyxf4f0RvnWMWZA4uCEdoeq_VE31s,3608
@@ -190,10 +190,10 @@ onnx2tf/utils/enums.py,sha256=7c5TqetqB07VjyHoxJHfLgtqBqk9ZRyUF33fPOJR1IM,1649
 onnx2tf/utils/iterative_json_optimizer.py,sha256=qqeIxWGxrhcCYk8-ebWnblnOkzDCwi-nseipHzHR_bk,10436
 onnx2tf/utils/json_auto_generator.py,sha256=C-4P8sii7B2_LRwW6AFG13tI3R5DHPhNirgxfmBU1F8,61944
 onnx2tf/utils/logging.py,sha256=yUCmPuJ_XiUItM3sZMcaMO24JErkQy7zZwVTYWAuiKg,1982
-onnx2tf-1.28.6.dist-info/licenses/LICENSE,sha256=5v_Kxihy8i6mzHVl349ikSREaIdsl9YeUnX1KBDLD2w,1070
-onnx2tf-1.28.6.dist-info/licenses/LICENSE_onnx-tensorflow,sha256=gK4GtS9S5YcyINu6uuNNWdo-kBClyEM4MFLFGiNTeRM,11231
-onnx2tf-1.28.6.dist-info/METADATA,sha256=w0m-5NPgED6SHFIOocSP1wySXbfaJDVal03zOHF4gQQ,151931
-onnx2tf-1.28.6.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-onnx2tf-1.28.6.dist-info/entry_points.txt,sha256=gDPK8ToCFPKMvm8jr9xrGOkXtORJJVh4736fBEKO5k0,41
-onnx2tf-1.28.6.dist-info/top_level.txt,sha256=WgfPiEy3f6vZ_FOpAIEA2CF3TCx1eYrhGw93Ih6b9Fw,8
-onnx2tf-1.28.6.dist-info/RECORD,,
+onnx2tf-1.28.8.dist-info/licenses/LICENSE,sha256=5v_Kxihy8i6mzHVl349ikSREaIdsl9YeUnX1KBDLD2w,1070
+onnx2tf-1.28.8.dist-info/licenses/LICENSE_onnx-tensorflow,sha256=gK4GtS9S5YcyINu6uuNNWdo-kBClyEM4MFLFGiNTeRM,11231
+onnx2tf-1.28.8.dist-info/METADATA,sha256=NKZj0tTAVIo80eNaKXJTxQDpRtl3-smFlcQCS3QpnMc,152862
+onnx2tf-1.28.8.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+onnx2tf-1.28.8.dist-info/entry_points.txt,sha256=gDPK8ToCFPKMvm8jr9xrGOkXtORJJVh4736fBEKO5k0,41
+onnx2tf-1.28.8.dist-info/top_level.txt,sha256=WgfPiEy3f6vZ_FOpAIEA2CF3TCx1eYrhGw93Ih6b9Fw,8
+onnx2tf-1.28.8.dist-info/RECORD,,