# Copyright 2020 - 2021 MONAI Consortium
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# http://www.apache.org/licenses/LICENSE-2.0
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import torch
import torch.nn as nn
from monai.utils import optional_import
einops, _ = optional_import("einops")
[docs]class SABlock(nn.Module):
"""
A self-attention block, based on: "Dosovitskiy et al.,
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale <https://arxiv.org/abs/2010.11929>"
"""
[docs] def __init__(self, hidden_size: int, num_heads: int, dropout_rate: float = 0.0) -> None:
"""
Args:
hidden_size: dimension of hidden layer.
num_heads: number of attention heads.
dropout_rate: faction of the input units to drop.
"""
super().__init__()
if not (0 <= dropout_rate <= 1):
raise ValueError("dropout_rate should be between 0 and 1.")
if hidden_size % num_heads != 0:
raise ValueError("hidden size should be divisible by num_heads.")
self.num_heads = num_heads
self.out_proj = nn.Linear(hidden_size, hidden_size)
self.qkv = nn.Linear(hidden_size, hidden_size * 3, bias=False)
self.drop_output = nn.Dropout(dropout_rate)
self.drop_weights = nn.Dropout(dropout_rate)
self.head_dim = hidden_size // num_heads
self.scale = self.head_dim ** -0.5
[docs] def forward(self, x):
q, k, v = einops.rearrange(self.qkv(x), "b h (qkv l d) -> qkv b l h d", qkv=3, l=self.num_heads)
att_mat = (torch.einsum("blxd,blyd->blxy", q, k) * self.scale).softmax(dim=-1)
att_mat = self.drop_weights(att_mat)
x = torch.einsum("bhxy,bhyd->bhxd", att_mat, v)
x = einops.rearrange(x, "b h l d -> b l (h d)")
x = self.out_proj(x)
x = self.drop_output(x)
return x