fix rvt bug

setup.py

@@ -3,7 +3,7 @@ from setuptools import setup, find_packages
 setup(
   name = 'vit-pytorch',
   packages = find_packages(exclude=['examples']),
-  version = '0.16.2',
+  version = '0.16.5',
   license='MIT',
   description = 'Vision Transformer (ViT) - Pytorch',
   author = 'Phil Wang',

vit_pytorch/rvt.py

@@ -19,7 +19,7 @@ class AxialRotaryEmbedding(nn.Module):
     def __init__(self, dim, max_freq = 10):
         super().__init__()
         self.dim = dim
-        scales = torch.logspace(1., log(max_freq / 2) / log(2), self.dim // 4, base = 2)
+        scales = torch.logspace(0., log(max_freq / 2) / log(2), self.dim // 4, base = 2)
         self.register_buffer('scales', scales)
 
     def forward(self, x):
@@ -94,7 +94,7 @@ class FeedForward(nn.Module):
         return self.net(x)
 
 class Attention(nn.Module):
-    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0., conv_query_kernel = 9):
+    def __init__(self, dim, heads = 8, dim_head = 64, dropout = 0., conv_query_kernel = 5):
         super().__init__()
         inner_dim = dim_head *  heads
 
@@ -192,4 +192,4 @@ class RvT(nn.Module):
         fmap_dims = {'h': h // p, 'w': w // p}
         x = self.transformer(x, fmap_dims = fmap_dims)
 
-        return self.mlp_head(x)
+        return self.mlp_head(x[:, 0])