fix bugs (#6240)

ppdet/modeling/backbones/__init__.py

@@ -32,6 +32,7 @@ from . import esnet
 from . import cspresnet
 from . import csp_darknet
 from . import convnext
+from . import vision_transformer
 
 from .vgg import *
 from .resnet import *
@@ -52,4 +53,5 @@ from .hardnet import *
 from .esnet import *
 from .cspresnet import *
 from .csp_darknet import *
-from .convnext import *
\ No newline at end of file
+from .convnext import *
+from .vision_transformer import *
\ No newline at end of file

ppdet/modeling/backbones/vision_transformer.py

@@ -347,6 +347,10 @@ class VisionTransformer(nn.Layer):
         self.embed_dim = embed_dim
         self.with_fpn = with_fpn
         self.use_checkpoint = use_checkpoint
+        self.use_sincos_pos_emb = use_sincos_pos_emb
+        self.use_rel_pos_bias = use_rel_pos_bias
+        self.final_norm = final_norm
+
         if use_checkpoint:
             print('please set: FLAGS_allocator_strategy=naive_best_fit')
         self.patch_embed = PatchEmbed(
@@ -369,8 +373,8 @@ class VisionTransformer(nn.Layer):
                     std=.02))
         elif use_sincos_pos_emb:
             pos_embed = self.build_2d_sincos_position_embedding(embed_dim)
-            self.pos_embed = pos_embed
 
+            self.pos_embed = pos_embed
             self.pos_embed = self.create_parameter(shape=pos_embed.shape)
             self.pos_embed.set_value(pos_embed.numpy())
             self.pos_embed.stop_gradient = True
@@ -383,15 +387,9 @@ class VisionTransformer(nn.Layer):
         if use_shared_rel_pos_bias:
             self.rel_pos_bias = RelativePositionBias(
                 window_size=self.patch_embed.patch_shape, num_heads=num_heads)
-
-        elif self.use_sincos_pos_emb:
-            self.pos_embed = self.build_2d_sincos_position_embedding(embed_dim)
-
         else:
             self.rel_pos_bias = None
 
-        self.use_rel_pos_bias = use_rel_pos_bias
-
         dpr = np.linspace(0, drop_path_rate, depth)
 
         self.blocks = nn.LayerList([
@@ -411,17 +409,15 @@ class VisionTransformer(nn.Layer):
                 epsilon=epsilon) for i in range(depth)
         ])
 
-        self.final_norm = final_norm
-        ######### del by xy
-        #if self.final_norm:
-        #    self.norm = eval(norm_layer)(embed_dim, epsilon=epsilon)
         self.pretrained = pretrained
         self.init_weight()
 
         assert len(out_indices) <= 4, ''
         self.out_indices = out_indices
         self.out_channels = [embed_dim for _ in range(len(out_indices))]
-        self.out_strides = [4, 8, 16, 32][-len(out_indices):]
+        self.out_strides = [4, 8, 16, 32][-len(out_indices):] if with_fpn else [
+            8 for _ in range(len(out_indices))
+        ]
 
         self.norm = Identity()