Removed sequence padding from ivy transpose convolutions as its resul…

…ts were conceptually incorrect. Performing transpose convolution given a sequence padding is a counter-approach to performing it based on given "same"/"valid" padding and output_shape. For now the former will only be available in the torch frontend.

ivy/functional/backends/jax/layers.py

@@ -74,7 +74,7 @@ def conv1d_transpose(
     x: JaxArray,
     filters: JaxArray,
     strides: Union[int, Tuple[int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -131,7 +131,7 @@ def conv2d_transpose(
     x: JaxArray,
     filters: JaxArray,
     strides: Union[int, Tuple[int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -217,7 +217,7 @@ def conv3d_transpose(
     x: JaxArray,
     filters: JaxArray,
     strides: Union[int, Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -322,7 +322,7 @@ def conv_general_transpose(
     x: JaxArray,
     filters: JaxArray,
     strides: Union[int, Tuple[int], Tuple[int, int], Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     dims: Optional[int] = 2,
@@ -342,9 +342,8 @@ def conv_general_transpose(
     filter_df = _get_filter_dataformat(dims)
     if data_format == "channel_first":
         x = jnp.transpose(x, (0, *range(2, dims + 2), 1))
-    x_shape = list(x.shape[1 : dims + 1])
     padding = _get_tranpose_padding(
-        x_shape, filters.shape, strides, padding, dims, dilations, output_shape
+        x.shape[1:], filters.shape, strides, padding, dims, dilations, output_shape
     )
     res = jnp.concatenate(
         [

ivy/functional/backends/numpy/layers.py

@@ -154,7 +154,7 @@ def conv1d_transpose(
     x: np.ndarray,
     filters: np.ndarray,
     strides: Union[int, Tuple[int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -234,7 +234,7 @@ def conv2d_transpose(
     x: np.ndarray,
     filters: np.ndarray,
     strides: Union[int, Tuple[int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -371,7 +371,7 @@ def conv3d_transpose(
     x: np.ndarray,
     filters: np.ndarray,
     strides: Union[int, Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -475,7 +475,7 @@ def conv_general_transpose(
     x: np.ndarray,
     filters: np.ndarray,
     strides: Union[int, Tuple[int], Tuple[int, int], Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     dims: Optional[int] = 2,

ivy/functional/backends/tensorflow/layers.py

@@ -90,7 +90,7 @@ def conv1d_transpose(
     x: Union[tf.Tensor, tf.Variable],
     filters: Union[tf.Tensor, tf.Variable],
     strides: Union[int, Tuple[int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -110,7 +110,6 @@ def conv1d_transpose(
     output_shape = _output_shape(
         x.shape, filters.shape, output_shape, strides, padding, 1, dilations
     )
-    padding = padding if isinstance(padding, str) else "VALID"
     res = tf.nn.conv1d_transpose(
         x, filters, output_shape, strides, padding, "NWC", dilations
     )
@@ -145,7 +144,7 @@ def conv2d_transpose(
     x: Union[tf.Tensor, tf.Variable],
     filters: Union[tf.Tensor, tf.Variable],
     strides: Union[int, Tuple[int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -165,7 +164,6 @@ def conv2d_transpose(
     output_shape = _output_shape(
         x.shape, filters.shape, output_shape, strides, padding, 2, dilations
     )
-    padding = padding if isinstance(padding, str) else "VALID"
     res = tf.nn.conv2d_transpose(
         x, filters, output_shape, strides, padding, "NHWC", dilations
     )
@@ -230,7 +228,7 @@ def conv3d_transpose(
     x: Tensor,
     filters: Tensor,
     strides: Union[int, Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -254,7 +252,6 @@ def conv3d_transpose(
     output_shape = _output_shape(
         x.shape, filters.shape, output_shape, strides[1:], padding, 3, dilations
     )
-    padding = padding if isinstance(padding, str) else "VALID"
     res = tf.nn.conv3d_transpose(
         x, filters, output_shape, strides, padding, "NDHWC", dilations
     )
@@ -377,7 +374,7 @@ def conv_general_transpose(
     x: Union[tf.Tensor, tf.Variable],
     filters: Union[tf.Tensor, tf.Variable],
     strides: Union[int, Tuple[int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     dims: Optional[int] = 2,

ivy/functional/backends/torch/layers.py

@@ -116,7 +116,7 @@ def conv1d_transpose(
     x: torch.Tensor,
     filters: torch.Tensor,
     strides: Union[int, Tuple[int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -128,10 +128,9 @@ def conv1d_transpose(
         x = x.permute(0, 2, 1)
     strides = [strides] if isinstance(strides, int) else strides
     dilations = [dilations] if isinstance(dilations, int) else dilations
-    filter_shape = list(filters.shape[0:1])
     filters = filters.permute(1, 2, 0)
     not_valid_pad, padding_list, output_padding = _pad_before_conv_tranpose(
-        x, filters, strides, padding, 1, dilations, output_shape, filter_shape
+        x, filters, strides, padding, 1, dilations, output_shape, filters.shape[2:]
     )
     res = torch.nn.functional.conv_transpose1d(
         x,
@@ -190,7 +189,7 @@ def conv2d_transpose(
     x: torch.Tensor,
     filters: torch.Tensor,
     strides: Union[int, Tuple[int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -202,10 +201,9 @@ def conv2d_transpose(
         x = x.permute(0, 3, 1, 2)
     strides = [strides] * 2 if isinstance(strides, int) else strides
     dilations = [dilations] * 2 if isinstance(dilations, int) else dilations
-    filter_shape = list(filters.shape[0:2])
     filters = filters.permute(2, 3, 0, 1)
     not_valid_pad, padding_list, output_padding = _pad_before_conv_tranpose(
-        x, filters, strides, padding, 2, dilations, output_shape, filter_shape
+        x, filters, strides, padding, 2, dilations, output_shape, filters.shape[2:]
     )
     res = torch.nn.functional.conv_transpose2d(
         x,
@@ -299,7 +297,7 @@ def conv3d_transpose(
     x: torch.Tensor,
     filters: torch.Tensor,
     strides: Union[int, Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.NativeShape, Sequence[int]]] = None,
@@ -311,10 +309,9 @@ def conv3d_transpose(
         x = x.permute(0, 4, 1, 2, 3)
     strides = [strides] * 3 if isinstance(strides, int) else strides
     dilations = [dilations] * 3 if isinstance(dilations, int) else dilations
-    filter_shape = list(filters.shape[0:3])
     filters = filters.permute(3, 4, 0, 1, 2)
     not_valid_pad, padding_list, output_padding = _pad_before_conv_tranpose(
-        x, filters, strides, padding, 3, dilations, output_shape, filter_shape
+        x, filters, strides, padding, 3, dilations, output_shape, filters.shape[2:]
     )
     res = torch.nn.functional.conv_transpose3d(
         x,
@@ -401,7 +398,7 @@ def conv_general_transpose(
     x: torch.Tensor,
     filters: torch.Tensor,
     strides: Union[int, Tuple[int], Tuple[int, int], Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     dims: Optional[int] = 2,
@@ -418,10 +415,9 @@ def conv_general_transpose(
         x = x.permute(0, dims + 1, *range(1, dims + 1))
     strides = [strides] * dims if isinstance(strides, int) else strides
     dilations = [dilations] * dims if isinstance(dilations, int) else dilations
-    filter_shape = list(filters.shape[0:dims])
     filters = filters.permute(dims, dims + 1, *range(dims))
     not_valid_pad, padding_list, output_padding = _pad_before_conv_tranpose(
-        x, filters, strides, padding, dims, dilations, output_shape, filter_shape
+        x, filters, strides, padding, dims, dilations, output_shape, filters.shape[2:]
     )
     if dims == 1:
         res = torch.nn.functional.conv_transpose1d(

ivy/functional/ivy/layers.py

@@ -884,7 +884,7 @@ def conv1d_transpose(
     x: Union[ivy.Array, ivy.NativeArray],
     filters: Union[ivy.Array, ivy.NativeArray],
     strides: Union[int, Tuple[int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.Shape, ivy.NativeShape]] = None,
@@ -903,9 +903,8 @@ def conv1d_transpose(
     strides
         The stride of the sliding window for each dimension of input.
     padding
-        either the string ‘SAME’ (padding with zeros evenly), the string ‘VALID’ (no
-        padding), or a sequence of n (low, high) integer pairs that give the padding to
-        apply before and after each spatial dimension.
+        Either ‘SAME’ (padding so that the output's shape is the same as the
+        input's), or ‘VALID’ (padding so that the output's shape is `output_shape`).
     output_shape
         Shape of the output (Default value = None)
     data_format
@@ -1145,7 +1144,7 @@ def conv2d_transpose(
     x: Union[ivy.Array, ivy.NativeArray],
     filters: Union[ivy.Array, ivy.NativeArray],
     strides: Union[int, Tuple[int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.Shape, ivy.NativeShape]] = None,
@@ -1164,9 +1163,8 @@ def conv2d_transpose(
     strides
         The stride of the sliding window for each dimension of input.
     padding
-        either the string ‘SAME’ (padding with zeros evenly), the string ‘VALID’ (no
-        padding), or a sequence of n (low, high) integer pairs that give the padding to
-        apply before and after each spatial dimension.
+        Either ‘SAME’ (padding so that the output's shape is the same as the
+        input's), or ‘VALID’ (padding so that the output's shape is `output_shape`).
     output_shape
         Shape of the output (Default value = None)
     data_format
@@ -1518,7 +1516,7 @@ def conv3d_transpose(
     x: Union[ivy.Array, ivy.NativeArray],
     filters: Union[ivy.Array, ivy.NativeArray],
     strides: Union[int, Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     output_shape: Optional[Union[ivy.Shape, ivy.NativeShape]] = None,
@@ -1537,9 +1535,8 @@ def conv3d_transpose(
     strides
         The stride of the sliding window for each dimension of input.
     padding
-        either the string ‘SAME’ (padding with zeros evenly), the string ‘VALID’ (no
-        padding), or a sequence of n (low, high) integer pairs that give the padding to
-        apply before and after each spatial dimension.
+        Either ‘SAME’ (padding so that the output's shape is the same as the
+        input's), or ‘VALID’ (padding so that the output's shape is `output_shape`).
     output_shape
         Shape of the output (Default value = None)
     data_format
@@ -1706,7 +1703,7 @@ def conv_general_transpose(
     x: Union[ivy.Array, ivy.NativeArray],
     filters: Union[ivy.Array, ivy.NativeArray],
     strides: Union[int, Tuple[int], Tuple[int, int], Tuple[int, int, int]],
-    padding: Union[str, Sequence[Tuple[int, int]]],
+    padding: str,
     /,
     *,
     dims: Optional[int] = 2,
@@ -1731,9 +1728,8 @@ def conv_general_transpose(
     strides
         The stride of the sliding window for each dimension of input.
     padding
-        either the string ‘SAME’ (padding with zeros evenly), the string ‘VALID’ (no
-        padding), or a sequence of n (low, high) integer pairs that give the padding to
-        apply before and after each spatial dimension.
+        Either ‘SAME’ (padding so that the output's shape is the same as the
+        input's), or ‘VALID’ (padding so that the output's shape is `output_shape`).
     dims
         Either 1, 2, or 3 corresponding to 1-D, 2-D, and 3-D convolution.
     output_shape

ivy_tests/test_ivy/test_functional/test_nn/test_layers.py

@@ -322,19 +322,6 @@ def x_and_filters(
 ):
     if not isinstance(dim, int):
         dim = draw(dim)
-    padding = draw(
-        st.one_of(
-            st.lists(
-                st.tuples(
-                    st.integers(min_value=0, max_value=3),
-                    st.integers(min_value=0, max_value=3),
-                ),
-                min_size=dim,
-                max_size=dim,
-            ),
-            st.sampled_from(["SAME", "VALID"]),
-        )
-    )
     batch_size = draw(st.integers(1, 5))
     filter_shape = draw(
         helpers.get_shape(
@@ -374,6 +361,7 @@ def x_and_filters(
     full_strides = [strides] * dim if isinstance(strides, int) else strides
     full_dilations = [dilations] * dim if isinstance(dilations, int) else dilations
     if transpose:
+        padding = draw(st.sampled_from(["SAME", "VALID"]))
         x_dim = draw(
             helpers.get_shape(
                 min_num_dims=dim, max_num_dims=dim, min_dim_size=1, max_dim_size=5
@@ -394,6 +382,19 @@ def x_and_filters(
         else:
             output_shape = None
     else:
+        padding = draw(
+            st.one_of(
+                st.lists(
+                    st.tuples(
+                        st.integers(min_value=0, max_value=3),
+                        st.integers(min_value=0, max_value=3),
+                    ),
+                    min_size=dim,
+                    max_size=dim,
+                ),
+                st.sampled_from(["SAME", "VALID"]),
+            )
+        )
         x_dim = []
         for i in range(dim):
             min_x = filter_shape[i] + (filter_shape[i] - 1) * (full_dilations[i] - 1)