use tensor.shape bug not paddle.shape(tensor) (#690)

* use tensor.shape bug not paddle.shape(tensor)

* refine

paddlevideo/modeling/backbones/swin_transformer.py

@@ -38,7 +38,7 @@ def drop_path(x, drop_prob=0., training=False):
     if drop_prob == 0. or not training:
         return x
     keep_prob = paddle.to_tensor(1 - drop_prob)
-    shape = (paddle.shape(x)[0], ) + (1, ) * (x.ndim - 1)
+    shape = (x.shape[0], ) + (1, ) * (x.ndim - 1)
     random_tensor = keep_prob + paddle.rand(shape, dtype=x.dtype)
     random_tensor = paddle.floor(random_tensor)  # binarize
     output = x.divide(keep_prob) * random_tensor
@@ -317,7 +317,7 @@ def __init__(self,
                        drop=drop)
 
     def forward_part1(self, x, mask_matrix):
-        B = paddle.shape(x)[0]
+        B = x.shape[0]
         _, D, H, W, C = x.shape
         window_size, shift_size = get_window_size((D, H, W), self.window_size,
                                                   self.shift_size)
@@ -513,7 +513,7 @@ def forward(self, x):
             x: Input feature, tensor size (B, C, D, H, W).
         """
         # calculate attention mask for SW-MSA
-        B = paddle.shape(x)[0]
+        B = x.shape[0]
         _, C, D, H, W = x.shape
         window_size, shift_size = get_window_size((D, H, W), self.window_size,
                                                   self.shift_size)

paddlevideo/modeling/backbones/toshift_vit.py

@@ -43,7 +43,7 @@ def drop_path(x, drop_prob=0., training=False):
     if drop_prob == 0. or not training:
         return x
     keep_prob = paddle.to_tensor(1 - drop_prob)
-    shape = (paddle.shape(x)[0], ) + (1, ) * (x.ndim - 1)
+    shape = (x.shape[0], ) + (1, ) * (x.ndim - 1)
     random_tensor = keep_prob + paddle.rand(shape).astype(x.dtype)
     random_tensor = paddle.floor(random_tensor)  # binarize
     output = x.divide(keep_prob) * random_tensor
@@ -374,7 +374,7 @@ def _init_fn(self, m):
 
     def forward_features(self, x):
         # B = x.shape[0]
-        B = paddle.shape(x)[0]
+        B = x.shape[0]
         x, T, W = self.patch_embed(x)  # [BT,nH*nW,F]
         cls_tokens = self.cls_token.expand((B * T, -1, -1))  # [1,1,F]->[BT,1,F]
         x = paddle.concat((cls_tokens, x), axis=1)

paddlevideo/modeling/backbones/vit.py

@@ -43,7 +43,7 @@ def drop_path(x, drop_prob=0., training=False):
     if drop_prob == 0. or not training:
         return x
     keep_prob = paddle.to_tensor(1 - drop_prob, dtype=x.dtype)
-    shape = (paddle.shape(x)[0], ) + (1, ) * (x.ndim - 1)
+    shape = (x.shape[0], ) + (1, ) * (x.ndim - 1)
     random_tensor = keep_prob + paddle.rand(shape).astype(x.dtype)
     random_tensor = paddle.floor(random_tensor)  # binarize
     output = x.divide(keep_prob) * random_tensor
@@ -394,7 +394,7 @@ def _init_fn(self, m):
 
     def forward_features(self, x):
         # B = x.shape[0]
-        B = paddle.shape(x)[0]
+        B = x.shape[0]
         x, T, W = self.patch_embed(x)  # [BT,nH*nW,F]
         cls_tokens = self.cls_token.expand((B * T, -1, -1))  # [1,1,F]->[BT,1,F]
         x = paddle.concat((cls_tokens, x), axis=1)

paddlevideo/modeling/backbones/vit_tweaks.py

@@ -65,7 +65,7 @@ def drop_path(x, drop_prob=0., training=False):
     if drop_prob == 0. or not training:
         return x
     keep_prob = paddle.to_tensor(1 - drop_prob)
-    shape = (paddle.shape(x)[0], ) + (1, ) * (x.ndim - 1)
+    shape = (x.shape[0], ) + (1, ) * (x.ndim - 1)
     random_tensor = keep_prob + paddle.rand(shape, dtype=x.dtype)
     random_tensor = paddle.floor(random_tensor)  # binarize
     output = x.divide(keep_prob) * random_tensor
@@ -444,7 +444,7 @@ def _init_fn(self, m):
 
     def forward_features(self, x):
         # B = x.shape[0]
-        B = paddle.shape(x)[0]
+        B = x.shape[0]
         x, T, W = self.patch_embed(x)  # [BT,nH*nW,F]
         cls_tokens = self.cls_token.expand((B * T, -1, -1))  # [1,1,F]->[BT,1,F]
         x = paddle.concat((cls_tokens, x), axis=1)

paddlevideo/modeling/heads/i3d_head.py

@@ -87,7 +87,7 @@ def forward(self, x):
         if self.dropout is not None:
             x = self.dropout(x)
         # [N, in_channels, 1, 1, 1]
-        N = paddle.shape(x)[0]
+        N = x.shape[0]
         x = x.reshape([N, -1])
         # [N, in_channels]
         cls_score = self.fc(x)