layernorm on channel dimension == instancenorm2d with affine set to true

README.md

@@ -378,6 +378,32 @@ img = torch.randn(1, 3, 224, 224)
 pred = model(img) # (1, 1000)
 ```
 
+## NesT
+
+<img src="./images/nest.png" width="400px"></img>
+
+This <a href="https://arxiv.org/abs/2105.12723">paper</a> decided to process the image in hierarchical stages, with attention only within tokens of local blocks, which aggregate as it moves up the heirarchy. The aggregation is done in the image plane, and contains a convolution and subsequent maxpool to allow it to pass information across the boundary.
+
+You can use it with the following code (ex. NesT-T)
+
+```python
+import torch
+from vit_pytorch.nest import NesT
+
+nest = NesT(
+    image_size = 224,
+    patch_size = 4,
+    dim = 96,
+    heads = 3,
+    num_hierarchies = 3,        # number of hierarchies
+    block_repeats = (8, 4, 1),  # the number of transformer blocks at each heirarchy, starting from the bottom
+    num_classes = 1000
+)
+
+img = torch.randn(1, 3, 224, 224)
+pred = nest(img) # (1, 1000)
+```
+
 ## Masked Patch Prediction
 
 Thanks to <a href="https://github.com/zankner">Zach</a>, you can train using the original masked patch prediction task presented in the paper, with the following code.
@@ -787,6 +813,17 @@ Coming from computer vision and new to transformers? Here are some resources tha
 }
 ```
 
+```bibtex
+@misc{zhang2021aggregating,
+    title   = {Aggregating Nested Transformers},
+    author  = {Zizhao Zhang and Han Zhang and Long Zhao and Ting Chen and Tomas Pfister},
+    year    = {2021},
+    eprint  = {2105.12723},
+    archivePrefix = {arXiv},
+    primaryClass = {cs.CV}
+}
+```
+
 ```bibtex
 @misc{caron2021emerging,
     title   = {Emerging Properties in Self-Supervised Vision Transformers},

setup.py

@@ -3,7 +3,7 @@ from setuptools import setup, find_packages
 setup(
   name = 'vit-pytorch',
   packages = find_packages(exclude=['examples']),
-  version = '0.18.4',
+  version = '0.19.4',
   license='MIT',
   description = 'Vision Transformer (ViT) - Pytorch',
   author = 'Phil Wang',

vit_pytorch/nest.py

@@ -0,0 +1,169 @@
+from functools import partial
+import torch
+from torch import nn, einsum
+
+from einops import rearrange
+from einops.layers.torch import Rearrange, Reduce
+
+# helpers
+
+def cast_tuple(val, depth):
+    return val if isinstance(val, tuple) else ((val,) * depth)
+
+LayerNorm = partial(nn.InstanceNorm2d, affine = True)
+
+# classes
+
+class PreNorm(nn.Module):
+    def __init__(self, dim, fn):
+        super().__init__()
+        self.norm = LayerNorm(dim)
+        self.fn = fn
+
+    def forward(self, x, **kwargs):
+        return self.fn(self.norm(x), **kwargs)
+
+class FeedForward(nn.Module):
+    def __init__(self, dim, mlp_mult = 4, dropout = 0.):
+        super().__init__()
+        self.net = nn.Sequential(
+            nn.Conv2d(dim, dim * mlp_mult, 1),
+            nn.GELU(),
+            nn.Dropout(dropout),
+            nn.Conv2d(dim * mlp_mult, dim, 1),
+            nn.Dropout(dropout)
+        )
+    def forward(self, x):
+        return self.net(x)
+
+class Attention(nn.Module):
+    def __init__(self, dim, heads = 8, dropout = 0.):
+        super().__init__()
+        dim_head = dim // heads
+        inner_dim = dim_head * heads
+        self.heads = heads
+        self.scale = dim_head ** -0.5
+
+        self.attend = nn.Softmax(dim = -1)
+        self.to_qkv = nn.Conv2d(dim, inner_dim * 3, 1, bias = False)
+
+        self.to_out = nn.Sequential(
+            nn.Conv2d(inner_dim, dim, 1),
+            nn.Dropout(dropout)
+        )
+
+    def forward(self, x):
+        b, c, h, w, heads = *x.shape, self.heads
+
+        qkv = self.to_qkv(x).chunk(3, dim = 1)
+        q, k, v = map(lambda t: rearrange(t, 'b (h d) x y -> b h (x y) d', h = heads), qkv)
+
+        dots = einsum('b h i d, b h j d -> b h i j', q, k) * self.scale
+
+        attn = self.attend(dots)
+
+        out = einsum('b h i j, b h j d -> b h i d', attn, v)
+        out = rearrange(out, 'b h (x y) d -> b (h d) x y', x = h, y = w)
+        return self.to_out(out)
+
+def Aggregate(dim, dim_out):
+    return nn.Sequential(
+        nn.Conv2d(dim, dim_out, 3, padding = 1),
+        LayerNorm(dim_out),
+        nn.MaxPool2d(3, stride = 2, padding = 1)
+    )
+
+class Transformer(nn.Module):
+    def __init__(self, dim, seq_len, depth, heads, mlp_mult, dropout = 0.):
+        super().__init__()
+        self.layers = nn.ModuleList([])
+        self.pos_emb = nn.Parameter(torch.randn(seq_len))
+
+        for _ in range(depth):
+            self.layers.append(nn.ModuleList([
+                PreNorm(dim, Attention(dim, heads = heads, dropout = dropout)),
+                PreNorm(dim, FeedForward(dim, mlp_mult, dropout = dropout))
+            ]))
+    def forward(self, x):
+        *_, h, w = x.shape
+
+        pos_emb = self.pos_emb[:(h * w)]
+        pos_emb = rearrange(pos_emb, '(h w) -> () () h w', h = h, w = w)
+        x = x + pos_emb
+
+        for attn, ff in self.layers:
+            x = attn(x) + x
+            x = ff(x) + x
+        return x
+
+class NesT(nn.Module):
+    def __init__(
+        self,
+        *,
+        image_size,
+        patch_size,
+        num_classes,
+        dim,
+        heads,
+        num_hierarchies,
+        block_repeats,
+        mlp_mult = 4,
+        channels = 3,
+        dim_head = 64,
+        dropout = 0.
+    ):
+        super().__init__()
+        assert (image_size % patch_size) == 0, 'Image dimensions must be divisible by the patch size.'
+        num_patches = (image_size // patch_size) ** 2
+        patch_dim = channels * patch_size ** 2
+        fmap_size = image_size // patch_size
+        blocks = 2 ** (num_hierarchies - 1)
+
+        seq_len = (fmap_size // blocks) ** 2   # sequence length is held constant across heirarchy
+        hierarchies = list(reversed(range(num_hierarchies)))
+        mults = [2 ** i for i in hierarchies]
+
+        layer_heads = list(map(lambda t: t * heads, mults))
+        layer_dims = list(map(lambda t: t * dim, mults))
+
+        layer_dims = [*layer_dims, layer_dims[-1]]
+        dim_pairs = zip(layer_dims[:-1], layer_dims[1:])
+
+        self.to_patch_embedding = nn.Sequential(
+            Rearrange('b c (h p1) (w p2) -> b (p1 p2 c) h w', p1 = patch_size, p2 = patch_size),
+            nn.Conv2d(patch_dim, layer_dims[0], 1),
+        )
+
+        block_repeats = cast_tuple(block_repeats, num_hierarchies)
+
+        self.layers = nn.ModuleList([])
+
+        for level, heads, (dim_in, dim_out), block_repeat in zip(hierarchies, layer_heads, dim_pairs, block_repeats):
+            is_last = level == 0
+            depth = block_repeat
+
+            self.layers.append(nn.ModuleList([
+                Transformer(dim_in, seq_len, depth, heads, mlp_mult, dropout),
+                Aggregate(dim_in, dim_out) if not is_last else nn.Identity()
+            ]))
+
+        self.mlp_head = nn.Sequential(
+            LayerNorm(dim),
+            Reduce('b c h w -> b c', 'mean'),
+            nn.Linear(dim, num_classes)
+        )
+
+    def forward(self, img):
+        x = self.to_patch_embedding(img)
+        b, c, h, w = x.shape
+
+        num_hierarchies = len(self.layers)
+
+        for level, (transformer, aggregate) in zip(reversed(range(num_hierarchies)), self.layers):
+            block_size = 2 ** level
+            x = rearrange(x, 'b c (b1 h) (b2 w) -> (b b1 b2) c h w', b1 = block_size, b2 = block_size)
+            x = transformer(x)
+            x = rearrange(x, '(b b1 b2) c h w -> b c (b1 h) (b2 w)', b1 = block_size, b2 = block_size)
+            x = aggregate(x)
+
+        return self.mlp_head(x)