fix resolution problem for swin transformer and clip vit (#3021)

* Update foundation_vit.py

Update .gitignore

fix time cost problem

Update swin_transformer.py

fix the speed and memory problem

reduce the unnecessary calculation when patch matches resolution

fix conflict

remove check resolution function

Revert "fix conflict"

This reverts commit d7a7dad.

fix conflict

remove the conflict checkpoint function

【Hackathon 5th No.69】 分类大模型--人体视觉任务SOLIDER (#2995)

* add_solider

* add_solider

* add_solider

* add_solider

* add_solider

* add_solider

* add_solider

* add_solider

* add_solider

* add_solider

* add_solider

update doc about PPHGNetV2 (#3002)

fix clip patch embedding resolution problem

support non 224 resolution

integrate the pading function to one

adjust function name

fix the resolution problem for clip-vision transformer part and swim transformer

fix the resolution problem for clip-vision transformer part and swim transformer

* fix cache problem

using the huggingface plan and drop the cache

* Revert "fix cache problem"

This reverts commit 8f7ab55.

* fix resolution problem

* update big model backbone

* Revert "update big model backbone"

This reverts commit 04a39f7.

ppcls/arch/backbone/legendary_models/swin_transformer.py

@@ -147,6 +147,8 @@ def pading_for_not_divisible(pixel_values,
                              function="split"):
     if isinstance(patch_size, int):
         patch_size = (patch_size, patch_size)
+    if height % patch_size[0] == 0 and width % patch_size[1] == 0:
+        return pixel_values, (0, 0, 0, 0, 0, 0, 0, 0)
     if function == "split":
         pading_width = patch_size[1] - width % patch_size[1]
         pading_height = patch_size[0] - height % patch_size[0]
@@ -407,7 +409,7 @@ def __init__(self,
                        act_layer=act_layer,
                        drop=drop)
         H, W = self.input_resolution
-        attn_mask = paddle.zeros([1, H, W, 1])
+        attn_mask = None
 
         self.register_buffer("attn_mask", attn_mask)
 
@@ -450,6 +452,9 @@ def forward(self, x, input_dimensions):
         x, pad_values = pading_for_not_divisible(x, H, W, self.window_size,
                                                  "BHWC")
         _, height_pad, width_pad, _ = x.shape
+
+        padding_state = pad_values[3] > 0 or pad_values[
+            5] > 0  # change variable name
         # cyclic shift
         if self.shift_size > 0:
             shifted_x = RollWrapper.roll(
@@ -465,7 +470,9 @@ def forward(self, x, input_dimensions):
              C])  # nW*B, window_size*window_size, C
 
         # W-MSA/SW-MSA
+        #check did it need to calculate again
         attn_mask = self.get_attn_mask(height_pad, width_pad, x.dtype)
+
         attn_windows = self.attn(
             x_windows, mask=attn_mask)  # nW*B, window_size*window_size, C
 
@@ -484,8 +491,7 @@ def forward(self, x, input_dimensions):
         else:
             x = shifted_x
 
-        was_padded = pad_values[3] > 0 or pad_values[5] > 0
-        if was_padded:
+        if padding_state:
             x = x[:, :H, :W, :]
         x = x.reshape([B, H * W, C])
 

ppcls/arch/backbone/model_zoo/foundation_vit.py

@@ -114,6 +114,8 @@ def pading_for_not_divisible(pixel_values,
                              function="split"):
     if isinstance(patch_size, int):
         patch_size = (patch_size, patch_size)
+    if height % patch_size[0] == 0 and width % patch_size[1] == 0:
+        return pixel_values, None
     if function == "split":
         pading_width = patch_size[1] - width % patch_size[1]
         pading_height = patch_size[0] - height % patch_size[0]