Image encoder
def window_partition(x: torch.Tensor, window_size: int) -> Tuple[torch.Tensor, Tuple[int, int]]:
"""
Partition into non-overlapping windows with padding if needed.
Args:
x (tensor): input tokens with [B, H, W, C].
window_size (int): window size.
Returns:
windows: windows after partition with [B * num_windows, window_size, window_size, C].
(Hp, Wp): padded height and width before partition
"""
B, H, W, C = x.shape
pad_h = (window_size - H % window_size) % window_size
pad_w = (window_size - W % window_size) % window_size
if pad_h > 0 or pad_w > 0:
x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h))
Hp, Wp = H + pad_h, W + pad_w
x = x.view(B, Hp // window_size, window_size, Wp // window_size, window_size, C)
windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
return windows, (Hp, Wp)def window_unpartition(
windows: torch.Tensor, window_size: int, pad_hw: Tuple[int, int], hw: Tuple[int, int]
) -> torch.Tensor:
"""
Window unpartition into original sequences and removing padding.
Args:
windows (tensor): input tokens with [B * num_windows, window_size, window_size, C].
window_size (int): window size.
pad_hw (Tuple): padded height and width (Hp, Wp).
hw (Tuple): original height and width (H, W) before padding.
Returns:
x: unpartitioned sequences with [B, H, W, C].
"""
Hp, Wp = pad_hw
H, W = hw
B = windows.shape[0] // (Hp * Wp // window_size // window_size)
x = windows.view(B, Hp // window_size, Wp // window_size, window_size, window_size, -1)
x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, Hp, Wp, -1)
if Hp > H or Wp > W:
x = x[:, :H, :W, :].contiguous()
return xdef get_rel_pos(q_size: int, k_size: int, rel_pos: torch.Tensor) -> torch.Tensor:
"""
Get relative positional embeddings according to the relative positions of
query and key sizes.
Args:
q_size (int): size of query q.
k_size (int): size of key k.
rel_pos (Tensor): relative position embeddings (L, C).
Returns:
Extracted positional embeddings according to relative positions.
"""
max_rel_dist = int(2 * max(q_size, k_size) - 1)
# Interpolate rel pos if needed.
# 这里interpolation 目的是如果inference时img size(或因为q,k feature scaling)不同 对原始rel_pos tensor缩放
# 而这个比例根据的是 max_rel_dist / rel_pos.shape[0]
#这里依据的是cross-atten between different layers(比如downsample产物)
#好处是聚焦算力 finer-atten & coarse-atten
#rel pos_emb 使得attend on multi scale 保留spatial reasoning
# 这里先将 [L, C] -> [1, C, L](一般格式) 然后 将L resize到max_dist(这里是选实际q,k feature map最长)
if rel_pos.shape[0] != max_rel_dist:
# Interpolate rel pos.
rel_pos_resized = F.interpolate(
rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1),
size=max_rel_dist,
mode="linear",
)
# ->[L_max, C]
rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).permute(1, 0)
else:
rel_pos_resized = rel_pos
# Scale the coords with short length if shapes for q and k are different.
q_coords = torch.arange(q_size)[:, None] * max(k_size / q_size, 1.0)
k_coords = torch.arange(k_size)[None, :] * max(q_size / k_size, 1.0)
relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)
return rel_pos_resized[relative_coords.long()]这里先看 rel_pos
"Input size must be provided if using relative positional encoding."
# input_size: Optional[Tuple[int, int]]
# initialize relative positional embeddings
self.rel_pos_h = nn.Parameter(torch.zeros(2 * input_size[0] - 1, head_dim))
self.rel_pos_w = nn.Parameter(torch.zeros(2 * input_size[1] - 1, head_dim))rel_pos可以学习 在h w两维各自长度为 2l -1 反应向上向下一共多少种可能性 每种可能性对应head_dim的emb info
这里的两方向 rel_pos + attn + q → 真正attn
attn = (q * self.scale) @ k.transpose(-2, -1)
if self.use_rel_pos:
attn = add_decomposed_rel_pos(attn, q, self.rel_pos_h, self.rel_pos_w, (H, W), (H, W))
attn = attn.softmax(dim=-1)
def add_decomposed_rel_pos(
attn: torch.Tensor,
q: torch.Tensor,
rel_pos_h: torch.Tensor,
rel_pos_w: torch.Tensor,
q_size: Tuple[int, int],
k_size: Tuple[int, int],
) -> torch.Tensor:
"""
Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py # noqa B950
Args:
attn (Tensor): attention map.
q (Tensor): query q in the attention layer with shape (B, q_h * q_w, C).
rel_pos_h (Tensor): relative position embeddings (Lh, C) for height axis.
rel_pos_w (Tensor): relative position embeddings (Lw, C) for width axis.
q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
k_size (Tuple): spatial sequence size of key k with (k_h, k_w).
Returns:
attn (Tensor): attention map with added relative positional embeddings.
"""
q_h, q_w = q_size
k_h, k_w = k_size
Rh = get_rel_pos(q_h, k_h, rel_pos_h)
Rw = get_rel_pos(q_w, k_w, rel_pos_w)
B, _, dim = q.shape
r_q = q.reshape(B, q_h, q_w, dim)
rel_h = torch.einsum("bhwc,hkc->bhwk", r_q, Rh)
rel_w = torch.einsum("bhwc,wkc->bhwk", r_q, Rw)
attn = (
attn.view(B, q_h, q_w, k_h, k_w) + rel_h[:, :, :, :, None] + rel_w[:, :, :, None, :]
).view(B, q_h * q_w, k_h * k_w)
return attn