recurrent_block.py

# Acknowledgement to
# https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/vision_transformer.py
import math
import logging
from functools import partial
from collections import OrderedDict
from copy import deepcopy
from itertools import repeat
import collections.abc

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch import _assert

from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD, IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD


# From PyTorch internals
def _ntuple(n):
    def parse(x):
        if isinstance(x, collections.abc.Iterable) and not isinstance(x, str):
            return tuple(x)
        return tuple(repeat(x, n))
    return parse

to_1tuple = _ntuple(1)
to_2tuple = _ntuple(2)
to_3tuple = _ntuple(3)
to_4tuple = _ntuple(4)
to_ntuple = _ntuple


class PatchEmbed(nn.Module):
    """ 2D Image to Patch Embedding
    """
    def __init__(
            self,
            img_size=224,
            patch_size=16,
            in_chans=3,
            embed_dim=768,
            norm_layer=None,
            flatten=True,
            bias=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, bias=bias)
        self.norm = norm_layer(embed_dim) if norm_layer else nn.Identity()

    def forward(self, x):
        B, C, H, W = x.shape
        _assert(H == self.img_size[0], f"Input image height ({H}) doesn't match model ({self.img_size[0]}).")
        _assert(W == self.img_size[1], f"Input image width ({W}) doesn't match model ({self.img_size[1]}).")
        x = self.proj(x)
        if self.flatten:
            x = x.flatten(2).transpose(1, 2)  # BCHW -> BNC
        x = self.norm(x)
        return x


class Mlp(nn.Module):
    """ MLP as used in Vision Transformer, MLP-Mixer and related networks
    """
    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, bias=True, drop=0.):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        bias = to_2tuple(bias)
        drop_probs = to_2tuple(drop)

        self.fc1 = nn.Linear(in_features, hidden_features, bias=bias[0])
        self.act = act_layer()
        self.drop1 = nn.Dropout(drop_probs[0])
        self.fc2 = nn.Linear(hidden_features, out_features, bias=bias[1])
        self.drop2 = nn.Dropout(drop_probs[1])

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.drop1(x)
        x = self.fc2(x)
        x = self.drop2(x)
        return x


class MHSC(nn.Module):
    def __init__(self, dim, num_heads=8, proj_bias=False, mhsc_drop=0., proj_drop=0.):
        super().__init__()
        self.num_heads = num_heads
        self.proj = nn.Linear(dim, dim, bias=proj_bias)
        self.mhsc_drop = nn.Dropout(mhsc_drop)
        self.proj_drop = nn.Dropout(proj_drop)

    def forward(self, x, mh_scm):
        B, K, D = x.shape
        mh_x = x.reshape(B, K, self.num_heads, D // self.num_heads).permute(0, 2, 1, 3)
        mh_scm = self.mhsc_drop(mh_scm)

        L = mh_scm.shape[-2]
        x = (mh_scm @ mh_x).transpose(1, 2).reshape(B, L, D)

        x = self.proj(x)
        x = self.proj_drop(x)

        return x


class Block(nn.Module):

    def __init__(self, dim, num_heads, mlp_ratio=4., proj_bias=False, drop=0., mhsc_drop=0.,
                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.mhsc = MHSC(dim, num_heads=num_heads, proj_bias=proj_bias, mhsc_drop=mhsc_drop, proj_drop=drop)
        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

    def forward(self, saet, scm, x):
        x = self.norm1(x + self.drop_path(self.mhsc(saet, scm)))
        x = self.norm2(x + self.drop_path(self.mlp(x)))

        return x


def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True):
    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
    This is the same as the DropConnect impl I created for EfficientNet, etc networks, however,
    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for
    changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use
    'survival rate' as the argument.
    """
    if drop_prob == 0. or not training:
        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)  # work with diff dim tensors, not just 2D ConvNets
    random_tensor = x.new_empty(shape).bernoulli_(keep_prob)
    if keep_prob > 0.0 and scale_by_keep:
        random_tensor.div_(keep_prob)
    return x * random_tensor


class DropPath(nn.Module):
    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks).
    """
    def __init__(self, drop_prob: float = 0., scale_by_keep: bool = True):
        super(DropPath, self).__init__()
        self.drop_prob = drop_prob
        self.scale_by_keep = scale_by_keep

    def forward(self, x):
        return drop_path(x, self.drop_prob, self.training, self.scale_by_keep)

    def extra_repr(self):
        return f'drop_prob={round(self.drop_prob,3):0.3f}'