onnx2tf 1.28.0__py3-none-any.whl → 1.28.2__py3-none-any.whl

onnx2tf/__init__.py

@@ -1,3 +1,3 @@
 from onnx2tf.onnx2tf import convert, main
 
-__version__ = '1.28.0'
+__version__ = '1.28.2'

onnx2tf/ops/MatMul.py

@@ -25,6 +25,59 @@ from onnx2tf.utils.enums import (
 )
 from typing import Any, Dict, List
 
+def _matmul_output_shape(shape_a, shape_b):
+    """ _matmul_output_shape
+    Computes the shape of matrix M=np.matmul(A, B) given shapes of A and B.
+    
+    Parameters
+    ----------
+        shape_a: tuple with shape of matrix A.
+        shape_b: tuple with shape of matrix B.
+
+    Returns
+    -------
+        Shape of matrix M=np.matmul(A, B)
+    """
+    if len(shape_a) == 0 or len(shape_b) == 0 or \
+            None in shape_a or None in shape_b or \
+            any(dim <= 0 for dim in shape_a) or any(dim <= 0 for dim in shape_b):
+        # If there are no dimensions, any dimension is None or not positive, we cannot determine the output shape
+        return None
+    # Handle 1D cases as per numpy.matmul rules
+    if len(shape_a) == 1 and len(shape_b) == 1:
+        # Vector dot product -> scalar
+        if shape_a[0] != shape_b[0]:
+            return None  # Incompatible shapes for matmul
+        else:
+            return ()
+    elif len(shape_a) == 1:
+        # (K,) @ (..., K, N) -> (..., N)
+        if shape_a[0] != shape_b[-2]:
+            return None  # Incompatible shapes for matmul
+        batch_shape = shape_b[:-2]
+        return batch_shape + (shape_b[-1],)
+    elif len(shape_b) == 1:
+        # (..., M, K) @ (K,) -> (..., M)
+        if shape_a[-1] != shape_b[0]:
+            return None  # Incompatible shapes for matmul
+        batch_shape = shape_a[:-2]
+        return batch_shape + (shape_a[-2],)
+    else:
+        # (..., M, K) @ (..., K, N) -> broadcast(...), M, N
+        # prepend the shorter shape with 1s to match lengths
+        if len(shape_a) < len(shape_b):
+            shape_a = (1,) * (len(shape_b) - len(shape_a)) + shape_a
+        elif len(shape_b) < len(shape_a):
+            shape_b = (1,) * (len(shape_a) - len(shape_b)) + shape_b
+
+        if shape_a[-1] != shape_b[-2]:
+            return None  # Incompatible shapes for matmul
+
+        try:
+            batch_shape = np.broadcast_shapes(shape_a[:-2], shape_b[:-2])
+        except ValueError:
+            return None  # If broadcasting fails, it means the batch dimensions are incompatible
+        return batch_shape + (shape_a[-2], shape_b[-1])
 
 @print_node_info
 @inverted_operation_enable_disable
@@ -199,8 +252,13 @@ def make_node(
                         and sum([1 if isinstance(s, str) else 0 for s in target_onnx_output_shape]) == 0:
                         dummy_np_1 = np.ones(list(input_tensor_1.shape), dtype=np.float32).transpose(tensor_1_candidate_for_transposition)
                         dummy_np_2 = np.ones(list(input_tensor_2.shape), dtype=np.float32).transpose(tensor_2_candidate_for_transposition)
-                        dummy_np_result: np.ndarray = np.matmul(dummy_np_1, dummy_np_2)
-                        if np.prod(dummy_np_result.shape) != np.prod(target_onnx_output_shape):
+
+                        actual_output_shape = _matmul_output_shape(dummy_np_1.shape, dummy_np_2.shape)
+                        if actual_output_shape is None:
+                            dummy_np_result: np.ndarray = np.matmul(dummy_np_1, dummy_np_2)
+                            actual_output_shape = dummy_np_result.shape
+                            
+                        if np.prod(actual_output_shape) != np.prod(target_onnx_output_shape):
                             continue
 
                     # Build TF dummy model

onnx2tf/utils/common_functions.py

@@ -97,7 +97,7 @@ def replace_parameter(
                     bool(replace_value) if isinstance(replace_value, int) and replace_value in [0, 1] else \
                     bool(int(replace_value)) if isinstance(replace_value, str) and replace_value in ["0", "1"] else \
                     False if isinstance(replace_value, str) and replace_value.lower() == "false" else \
-                    True if isinstance(replace_value, str) and replace_value.lower() == "True" else \
+                    True if isinstance(replace_value, str) and replace_value.lower() == "true" else \
                     replace_value
             elif isinstance(value_before_replacement, int):
                 replace_value = int(replace_value)

{onnx2tf-1.28.0.dist-info → onnx2tf-1.28.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: onnx2tf
-Version: 1.28.0
+Version: 1.28.2
 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:1.28.0
+  ghcr.io/pinto0309/onnx2tf:1.28.2
 
   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:1.28.0
+  docker.io/pinto0309/onnx2tf:1.28.2
 
   or
 

{onnx2tf-1.28.0.dist-info → onnx2tf-1.28.2.dist-info}/RECORD

@@ -1,4 +1,4 @@
-onnx2tf/__init__.py,sha256=_bz5n8zEp9SjBSnI_J1s35oR-5bngGBMjVCVYVDdQ54,66
+onnx2tf/__init__.py,sha256=Q-EJ5Kmj8v4pyqNo39_LSMw82TTyA_BlTLfHEgfsSiE,66
 onnx2tf/__main__.py,sha256=2RSCQ7d4lc6CwD-rlGn9UicPFg-P5du7ZD_yh-kuBEU,57
 onnx2tf/onnx2tf.py,sha256=ufjdjeokS96PyvhLAV4nOaiPZ69FCP2kbvBVGahzxxQ,146784
 onnx2tf/ops/Abs.py,sha256=V7btmCG_ZvK_qJovUsguq0ZMJ349mhNQ4FHSgzP_Yuo,4029
@@ -82,7 +82,7 @@ onnx2tf/ops/LessOrEqual.py,sha256=9Lc8qaYUPVC6yZoQluNqcdHnvpUbfWBOI4Ow38RRAJo,45
 onnx2tf/ops/Log.py,sha256=UZebF3SGq85BnoPgYyN2j-zzFRp67fJnYPNyu33W55o,3582
 onnx2tf/ops/LogSoftmax.py,sha256=j2nhYY7__8ViLFJVLA5tS98QEvGS1gTIW0QCdnZWUPQ,3923
 onnx2tf/ops/LpNormalization.py,sha256=Uu15HgxFNXb6gNMgdTJyf0SLPaLbcbkOYqY_4hMBxNA,3153
-onnx2tf/ops/MatMul.py,sha256=95HrWr3Dt6BLqx_zqm3WXBw_WzrWLObYVgz4K1yrhqE,19060
+onnx2tf/ops/MatMul.py,sha256=KHhRyQCyxe6845f-AOI1UJzA3rGTssG6eyKmDw0oegs,21466
 onnx2tf/ops/MatMulInteger.py,sha256=qHqzdJNI9SeJDbW8pR90baYCdGN6FdOez4hi9EzwXoc,6538
 onnx2tf/ops/Max.py,sha256=w5nMciO_6ApYUobHuwMGuS3xhuza7eSvKDRhvMPgAuo,3256
 onnx2tf/ops/MaxPool.py,sha256=_JC4eqBTh-qLkZCMG8RZhthRZ8D2d821zaFMWeGMEWc,15775
@@ -185,15 +185,15 @@ onnx2tf/ops/_Loop.py,sha256=eo5sNfrfOnKV6_I737AWsM5LJTY9DVOxQEvhanxtP4g,11322
 onnx2tf/ops/__Loop.py,sha256=ClwMcbNS4hqUtW_pzwjMa9Cqg7ONvz9aplke55A0uJ0,19704
 onnx2tf/ops/__init__.py,sha256=jnmUWWa-3dHzBZV9bmPzXu6eoz2dumJTzO7i8JdcgSM,25
 onnx2tf/utils/__init__.py,sha256=E9FM9He68VIASDnYp-OrxvHFVn55GzWqw2OEkCqn1zg,27
-onnx2tf/utils/common_functions.py,sha256=QdIh2V8st7R-7H6pyOtTb3oOgNeFZDX7YH21H63_XDA,245602
+onnx2tf/utils/common_functions.py,sha256=jdDz38sZdi2pB8QQKNSMbOb8w_Ju-RAqoScJ_-2DyNk,245602
 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=Vyy21SYEoSL0b-I1cUnaXR-CPoO8LJYQ3fAS2ulZSMM,61964
 onnx2tf/utils/logging.py,sha256=yUCmPuJ_XiUItM3sZMcaMO24JErkQy7zZwVTYWAuiKg,1982
-onnx2tf-1.28.0.dist-info/licenses/LICENSE,sha256=5v_Kxihy8i6mzHVl349ikSREaIdsl9YeUnX1KBDLD2w,1070
-onnx2tf-1.28.0.dist-info/licenses/LICENSE_onnx-tensorflow,sha256=gK4GtS9S5YcyINu6uuNNWdo-kBClyEM4MFLFGiNTeRM,11231
-onnx2tf-1.28.0.dist-info/METADATA,sha256=AAAKjgJIdmg6sMeDYnhWeA0aUDTl1-2PB5U-vFPMKas,151177
-onnx2tf-1.28.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-onnx2tf-1.28.0.dist-info/entry_points.txt,sha256=gDPK8ToCFPKMvm8jr9xrGOkXtORJJVh4736fBEKO5k0,41
-onnx2tf-1.28.0.dist-info/top_level.txt,sha256=WgfPiEy3f6vZ_FOpAIEA2CF3TCx1eYrhGw93Ih6b9Fw,8
-onnx2tf-1.28.0.dist-info/RECORD,,
+onnx2tf-1.28.2.dist-info/licenses/LICENSE,sha256=5v_Kxihy8i6mzHVl349ikSREaIdsl9YeUnX1KBDLD2w,1070
+onnx2tf-1.28.2.dist-info/licenses/LICENSE_onnx-tensorflow,sha256=gK4GtS9S5YcyINu6uuNNWdo-kBClyEM4MFLFGiNTeRM,11231
+onnx2tf-1.28.2.dist-info/METADATA,sha256=psCTn2IAkkXxmDaFe_1wLdA_lyCdP_ooOUvoiOluBfk,151177
+onnx2tf-1.28.2.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+onnx2tf-1.28.2.dist-info/entry_points.txt,sha256=gDPK8ToCFPKMvm8jr9xrGOkXtORJJVh4736fBEKO5k0,41
+onnx2tf-1.28.2.dist-info/top_level.txt,sha256=WgfPiEy3f6vZ_FOpAIEA2CF3TCx1eYrhGw93Ih6b9Fw,8
+onnx2tf-1.28.2.dist-info/RECORD,,