GLEE/glee/backbone/vit_utils.py

# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
import math
import numpy as np
from scipy import interpolate
import torch
import torch.nn as nn
import torch.nn.functional as F

__all__ = [
    "window_partition",
    "window_unpartition",
    "add_decomposed_rel_pos",
    "get_abs_pos",
    "PatchEmbed",
]


def window_partition(x, window_size):
    """
    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, window_size, pad_hw, hw):
    """
    Window unpartition into original sequences and removing padding.
    Args:
        x (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 x


def get_rel_pos(q_size, k_size, rel_pos, interp_type):
    """
    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.
    if rel_pos.shape[0] != max_rel_dist:
        if interp_type == "vitdet":
            # the vitdet impl: 
            # https://github.com/facebookresearch/detectron2/blob/96c752ce821a3340e27edd51c28a00665dd32a30/detectron2/modeling/backbone/utils.py#L77.

            rel_pos_resized = F.interpolate(
                rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1),
                size=max_rel_dist,
                mode="linear",
            )
            rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).permute(1, 0)
        elif interp_type == "beit":
            # steal from beit https://github.com/microsoft/unilm/tree/master/beit
            # modified by Yuxin Fang

            src_size = rel_pos.shape[0]
            dst_size = max_rel_dist

            q = 1.0903078
            dis = []

            cur = 1
            for i in range(src_size // 2):
                dis.append(cur)
                cur += q ** (i + 1)

            r_ids = [-_ for _ in reversed(dis)]
            x = r_ids + [0] + dis
            t = dst_size // 2.0
            dx = np.arange(-t, t + 0.1, 1.0)

            all_rel_pos_bias = []
            for i in range(rel_pos.shape[1]):
                # a hack from https://github.com/baaivision/EVA/issues/8,
                # could also be used in fine-tuning but the performance haven't been tested.
                z = rel_pos[:, i].view(src_size).cpu().float().detach().numpy()
                f = interpolate.interp1d(x, z, kind='cubic', fill_value="extrapolate")
                all_rel_pos_bias.append(
                    torch.Tensor(f(dx)).contiguous().view(-1, 1).to(rel_pos.device))
            rel_pos_resized = torch.cat(all_rel_pos_bias, dim=-1)
        else:
            raise NotImplementedError()
    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()]


def add_decomposed_rel_pos(attn, q, rel_pos_h, rel_pos_w, q_size, k_size, interp_type):
    """
    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, interp_type)
    Rw = get_rel_pos(q_w, k_w, rel_pos_w, interp_type)

    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


def get_abs_pos(abs_pos, has_cls_token, hw):
    """
    Calculate absolute positional embeddings. If needed, resize embeddings and remove cls_token
        dimension for the original embeddings.
    Args:
        abs_pos (Tensor): absolute positional embeddings with (1, num_position, C).
        has_cls_token (bool): If true, has 1 embedding in abs_pos for cls token.
        hw (Tuple): size of input image tokens.

    Returns:
        Absolute positional embeddings after processing with shape (1, H, W, C)
    """
    h, w = hw
    if has_cls_token:
        abs_pos = abs_pos[:, 1:]
    xy_num = abs_pos.shape[1]
    size = int(math.sqrt(xy_num))
    assert size * size == xy_num

    if size != h or size != w:
        new_abs_pos = F.interpolate(
            abs_pos.reshape(1, size, size, -1).permute(0, 3, 1, 2),
            size=(h, w),
            mode="bicubic",
            align_corners=False,
        )

        return new_abs_pos.permute(0, 2, 3, 1)
    else:
        return abs_pos.reshape(1, h, w, -1)


class PatchEmbed(nn.Module):
    """
    Image to Patch Embedding.
    """

    def __init__(
        self, kernel_size=(16, 16), stride=(16, 16), padding=(0, 0), in_chans=3, embed_dim=768
    ):
        """
        Args:
            kernel_size (Tuple): kernel size of the projection layer.
            stride (Tuple): stride of the projection layer.
            padding (Tuple): padding size of the projection layer.
            in_chans (int): Number of input image channels.
            embed_dim (int):  embed_dim (int): Patch embedding dimension.
        """
        super().__init__()

        self.proj = nn.Conv2d(
            in_chans, embed_dim, kernel_size=kernel_size, stride=stride, padding=padding
        )

    def forward(self, x):
        x = self.proj(x)
        # B C H W -> B H W C
        x = x.permute(0, 2, 3, 1)
        return x