Spaces:
Running
on
Zero
Running
on
Zero
File size: 5,177 Bytes
c8b42eb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 |
# --------------------------------------------------------
# Patch Embedding for CroCo and DUSt3R
# Adopted from DUSt3R (Naver Corporation, CC BY-NC-SA 4.0 (non-commercial use only))
# --------------------------------------------------------
import torch
import torch.nn as nn
from uniception.models.libs.croco.blocks import to_2tuple
torch.backends.cuda.matmul.allow_tf32 = True # for gpu >= Ampere and pytorch >= 1.12
def get_patch_embed(patch_embed_cls, img_size, patch_size, enc_embed_dim):
assert patch_embed_cls in ["PatchEmbedCroCo", "PatchEmbedDust3R", "ManyAR_PatchEmbed"]
patch_embed = eval(patch_embed_cls)(img_size, patch_size, 3, enc_embed_dim)
return patch_embed
class PositionGetter(object):
"""Return positions of patches"""
def __init__(self):
self.cache_positions = {}
def __call__(self, b, h, w, device):
if not (h, w) in self.cache_positions:
x = torch.arange(w, device=device)
y = torch.arange(h, device=device)
self.cache_positions[h, w] = torch.cartesian_prod(y, x) # (h, w, 2)
pos = self.cache_positions[h, w].view(1, h * w, 2).expand(b, -1, 2).clone()
return pos
class PatchEmbedCroCo(nn.Module):
"""Just adding _init_weights + position getter compared to timm.models.layers.patch_embed.PatchEmbed"""
def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, norm_layer=None, flatten=True):
super().__init__()
img_size = to_2tuple(img_size)
patch_size = to_2tuple(patch_size)
self.img_size = img_size
self.patch_size = patch_size
self.grid_size = (img_size[0] // patch_size[0], img_size[1] // patch_size[1])
self.num_patches = self.grid_size[0] * self.grid_size[1]
self.flatten = flatten
self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
self.norm = norm_layer(embed_dim) if norm_layer else nn.Identity()
self.position_getter = PositionGetter()
def forward(self, x, **kw):
B, C, H, W = x.shape
torch._assert(H == self.img_size[0], f"Input image height ({H}) doesn't match model ({self.img_size[0]}).")
torch._assert(W == self.img_size[1], f"Input image width ({W}) doesn't match model ({self.img_size[1]}).")
x = self.proj(x)
pos = self.position_getter(B, x.size(2), x.size(3), x.device)
if self.flatten:
x = x.flatten(2).transpose(1, 2) # BCHW -> BNC
x = self.norm(x)
return x, pos
def _init_weights(self):
w = self.proj.weight.data
torch.nn.init.xavier_uniform_(w.view([w.shape[0], -1]))
class PatchEmbedDust3R(PatchEmbedCroCo):
def forward(self, x, **kw):
B, C, H, W = x.shape
assert (
H % self.patch_size[0] == 0
), f"Input image height ({H}) is not a multiple of patch size ({self.patch_size[0]})."
assert (
W % self.patch_size[1] == 0
), f"Input image width ({W}) is not a multiple of patch size ({self.patch_size[1]})."
x = self.proj(x)
pos = self.position_getter(B, x.size(2), x.size(3), x.device)
if self.flatten:
x = x.flatten(2).transpose(1, 2) # BCHW -> BNC
x = self.norm(x)
return x, pos
class ManyAR_PatchEmbed(PatchEmbedCroCo):
"""Handle images with non-square aspect ratio.
All images in the same batch have the same aspect ratio.
true_shape = [(height, width) ...] indicates the actual shape of each image.
"""
def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, norm_layer=None, flatten=True):
self.embed_dim = embed_dim
super().__init__(img_size, patch_size, in_chans, embed_dim, norm_layer, flatten)
def forward(self, img, true_shape):
B, C, H, W = img.shape
assert W >= H, f"img should be in landscape mode, but got {W=} {H=}"
assert (
H % self.patch_size[0] == 0
), f"Input image height ({H}) is not a multiple of patch size ({self.patch_size[0]})."
assert (
W % self.patch_size[1] == 0
), f"Input image width ({W}) is not a multiple of patch size ({self.patch_size[1]})."
assert true_shape.shape == (B, 2), f"true_shape has the wrong shape={true_shape.shape}"
# size expressed in tokens
W //= self.patch_size[0]
H //= self.patch_size[1]
n_tokens = H * W
height, width = true_shape.T
is_landscape = width >= height
is_portrait = ~is_landscape
# allocate result
x = img.new_zeros((B, n_tokens, self.embed_dim))
pos = img.new_zeros((B, n_tokens, 2), dtype=torch.int64)
# linear projection, transposed if necessary
x[is_landscape] = self.proj(img[is_landscape]).permute(0, 2, 3, 1).flatten(1, 2).float()
x[is_portrait] = self.proj(img[is_portrait].swapaxes(-1, -2)).permute(0, 2, 3, 1).flatten(1, 2).float()
pos[is_landscape] = self.position_getter(1, H, W, pos.device)
pos[is_portrait] = self.position_getter(1, W, H, pos.device)
x = self.norm(x)
return x, pos
|