ann.py

# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# 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 os

import paddle
import paddle.nn.functional as F
from paddle import nn

from paddleseg.cvlibs import manager
from paddleseg.models.common import layer_libs
from paddleseg.utils import utils


@manager.MODELS.add_component
class ANN(nn.Layer):
    """
    The ANN implementation based on PaddlePaddle.

    The original article refers to 
        Zhen, Zhu, et al. "Asymmetric Non-local Neural Networks for Semantic Segmentation."
        (https://arxiv.org/pdf/1908.07678.pdf)

    It mainly consists of AFNB and APNB modules.

    Args:
        num_classes (int): the unique number of target classes.
        backbone (Paddle.nn.Layer): backbone network, currently support Resnet50/101.
        model_pretrained (str): the path of pretrained model. Default to None.
        backbone_indices (tuple): two values in the tuple indicate the indices of output of backbone.
            the first index will be taken as low-level features; the second one will be 
            taken as high-level features in AFNB module. Usually backbone consists of four 
            downsampling stage, and return an output of each stage, so we set default (2, 3), 
            which means taking feature map of the third stage and the fourth stage in backbone.
        backbone_channels (tuple): the same length with "backbone_indices". It indicates the channels of corresponding index.
        key_value_channels (int): the key and value channels of self-attention map in both AFNB and APNB modules.
            Default to 256.
        inter_channels (int): both input and output channels of APNB modules.
        psp_size (tuple): the out size of pooled feature maps. Default to (1, 3, 6, 8).
        enable_auxiliary_loss (bool): a bool values indicates whether adding auxiliary loss. Default to True.
    """

    def __init__(self,
                 num_classes,
                 backbone,
                 model_pretrained=None,
                 backbone_indices=(2, 3),
                 backbone_channels=(1024, 2048),
                 key_value_channels=256,
                 inter_channels=512,
                 psp_size=(1, 3, 6, 8),
                 enable_auxiliary_loss=True):
        super(ANN, self).__init__()

        self.backbone = backbone

        low_in_channels = backbone_channels[0]
        high_in_channels = backbone_channels[1]

        self.fusion = AFNB(
            low_in_channels=low_in_channels,
            high_in_channels=high_in_channels,
            out_channels=high_in_channels,
            key_channels=key_value_channels,
            value_channels=key_value_channels,
            dropout_prob=0.05,
            sizes=([1]),
            psp_size=psp_size)

        self.context = nn.Sequential(
            layer_libs.ConvBNReLU(
                in_channels=high_in_channels,
                out_channels=inter_channels,
                kernel_size=3,
                padding=1),
            APNB(
                in_channels=inter_channels,
                out_channels=inter_channels,
                key_channels=key_value_channels,
                value_channels=key_value_channels,
                dropout_prob=0.05,
                sizes=([1]),
                psp_size=psp_size))

        self.cls = nn.Conv2d(
            in_channels=inter_channels, out_channels=num_classes, kernel_size=1)
        self.auxlayer = layer_libs.AuxLayer(
            in_channels=low_in_channels,
            inter_channels=low_in_channels // 2,
            out_channels=num_classes,
            dropout_prob=0.05)

        self.backbone_indices = backbone_indices
        self.enable_auxiliary_loss = enable_auxiliary_loss

        self.init_weight(model_pretrained)

    def forward(self, input, label=None):

        logit_list = []
        _, feat_list = self.backbone(input)
        low_level_x = feat_list[self.backbone_indices[0]]
        high_level_x = feat_list[self.backbone_indices[1]]
        x = self.fusion(low_level_x, high_level_x)
        x = self.context(x)
        logit = self.cls(x)
        logit = F.resize_bilinear(logit, input.shape[2:])
        logit_list.append(logit)

        if self.enable_auxiliary_loss:
            auxiliary_logit = self.auxlayer(low_level_x)
            auxiliary_logit = F.resize_bilinear(auxiliary_logit,
                                                input.shape[2:])
            logit_list.append(auxiliary_logit)

        return logit_list

    def init_weight(self, pretrained_model=None):
        """
        Initialize the parameters of model parts.

        Args:
            pretrained_model ([str], optional): the pretrained_model path of backbone. Defaults to None.
        """

        if pretrained_model is not None:
            if os.path.exists(pretrained_model):
                utils.load_pretrained_model(self.backbone, pretrained_model)


class AFNB(nn.Layer):
    """
    Asymmetric Fusion Non-local Block

    Args:
        low_in_channels (int): low-level-feature channels.
        high_in_channels (int): high-level-feature channels.
        out_channels (int): out channels of AFNB module.
        key_channels (int): the key channels in self-attention block.
        value_channels (int): the value channels in self-attention block.
        dropout_prob (float): the dropout rate of output.
        sizes (tuple): the number of AFNB modules. Default to ([1]).
        psp_size (tuple): the out size of pooled feature maps. Default to (1, 3, 6, 8).
    """

    def __init__(self,
                 low_in_channels,
                 high_in_channels,
                 out_channels,
                 key_channels,
                 value_channels,
                 dropout_prob,
                 sizes=([1]),
                 psp_size=(1, 3, 6, 8)):
        super(AFNB, self).__init__()

        self.psp_size = psp_size
        self.stages = nn.LayerList([
            SelfAttentionBlock_AFNB(low_in_channels, high_in_channels,
                                    key_channels, value_channels, out_channels,
                                    size) for size in sizes
        ])
        self.conv_bn = layer_libs.ConvBn(
            in_channels=out_channels + high_in_channels,
            out_channels=out_channels,
            kernel_size=1)
        self.dropout_prob = dropout_prob

    def forward(self, low_feats, high_feats):
        priors = [stage(low_feats, high_feats) for stage in self.stages]
        context = priors[0]
        for i in range(1, len(priors)):
            context += priors[i]

        output = self.conv_bn(paddle.concat([context, high_feats], axis=1))
        output = F.dropout(output, p=self.dropout_prob)  # dropout_prob

        return output


class APNB(nn.Layer):
    """
    Asymmetric Pyramid Non-local Block

    Args:
        in_channels (int): the input channels of APNB module.
        out_channels (int): out channels of APNB module.
        key_channels (int): the key channels in self-attention block.
        value_channels (int): the value channels in self-attention block.
        dropout_prob (float): the dropout rate of output.
        sizes (tuple): the number of AFNB modules. Default to ([1]).
        psp_size (tuple): the out size of pooled feature maps. Default to (1, 3, 6, 8).
    """

    def __init__(self,
                 in_channels,
                 out_channels,
                 key_channels,
                 value_channels,
                 dropout_prob,
                 sizes=([1]),
                 psp_size=(1, 3, 6, 8)):
        super(APNB, self).__init__()

        self.psp_size = psp_size
        self.stages = nn.LayerList([
            SelfAttentionBlock_APNB(in_channels, out_channels, key_channels,
                                    value_channels, size) for size in sizes
        ])
        self.conv_bn = layer_libs.ConvBNReLU(
            in_channels=in_channels * 2,
            out_channels=out_channels,
            kernel_size=1)
        self.dropout_prob = dropout_prob

    def forward(self, feats):
        priors = [stage(feats) for stage in self.stages]
        context = priors[0]
        for i in range(1, len(priors)):
            context += priors[i]

        output = self.conv_bn(paddle.concat([context, feats], axis=1))
        output = F.dropout(output, p=self.dropout_prob)  # dropout_prob

        return output


def _pp_module(x, psp_size):
    n, c, h, w = x.shape
    priors = []
    for size in psp_size:
        feat = F.adaptive_pool2d(x, pool_size=size, pool_type="avg")
        feat = paddle.reshape(feat, shape=(n, c, -1))
        priors.append(feat)
    center = paddle.concat(priors, axis=-1)
    return center


class SelfAttentionBlock_AFNB(nn.Layer):
    """
    Self-Attention Block for AFNB module.

    Args:
        low_in_channels (int): low-level-feature channels.
        high_in_channels (int): high-level-feature channels.
        key_channels (int): the key channels in self-attention block.
        value_channels (int): the value channels in self-attention block.
        out_channels (int): out channels of AFNB module.
        scale (int): pooling size. Defaut to 1.
        psp_size (tuple): the out size of pooled feature maps. Default to (1, 3, 6, 8).
    """

    def __init__(self,
                 low_in_channels,
                 high_in_channels,
                 key_channels,
                 value_channels,
                 out_channels=None,
                 scale=1,
                 psp_size=(1, 3, 6, 8)):
        super(SelfAttentionBlock_AFNB, self).__init__()

        self.scale = scale
        self.in_channels = low_in_channels
        self.out_channels = out_channels
        self.key_channels = key_channels
        self.value_channels = value_channels
        if out_channels == None:
            self.out_channels = high_in_channels
        self.pool = nn.Pool2D(pool_size=(scale, scale), pool_type="max")
        self.f_key = layer_libs.ConvBNReLU(
            in_channels=low_in_channels,
            out_channels=key_channels,
            kernel_size=1)
        self.f_query = layer_libs.ConvBNReLU(
            in_channels=high_in_channels,
            out_channels=key_channels,
            kernel_size=1)
        self.f_value = nn.Conv2d(
            in_channels=low_in_channels,
            out_channels=value_channels,
            kernel_size=1)

        self.W = nn.Conv2d(
            in_channels=value_channels,
            out_channels=out_channels,
            kernel_size=1)

        self.psp_size = psp_size

    def forward(self, low_feats, high_feats):
        batch_size, _, h, w = high_feats.shape

        value = self.f_value(low_feats)
        value = _pp_module(value, self.psp_size)
        value = paddle.transpose(value, (0, 2, 1))

        query = self.f_query(high_feats)
        query = paddle.reshape(query, shape=(batch_size, self.key_channels, -1))
        query = paddle.transpose(query, perm=(0, 2, 1))

        key = self.f_key(low_feats)
        key = _pp_module(key, self.psp_size)

        sim_map = paddle.matmul(query, key)
        sim_map = (self.key_channels**-.5) * sim_map
        sim_map = F.softmax(sim_map, axis=-1)

        context = paddle.matmul(sim_map, value)
        context = paddle.transpose(context, perm=(0, 2, 1))
        context = paddle.reshape(
            context,
            shape=[batch_size, self.value_channels, *high_feats.shape[2:]])

        context = self.W(context)

        return context


class SelfAttentionBlock_APNB(nn.Layer):
    """
    Self-Attention Block for APNB module.

    Args:
        in_channels (int): the input channels of APNB module.
        out_channels (int): out channels of APNB module.
        key_channels (int): the key channels in self-attention block.
        value_channels (int): the value channels in self-attention block.
        scale (int): pooling size. Defaut to 1.
        psp_size (tuple): the out size of pooled feature maps. Default to (1, 3, 6, 8).
    """

    def __init__(self,
                 in_channels,
                 out_channels,
                 key_channels,
                 value_channels,
                 scale=1,
                 psp_size=(1, 3, 6, 8)):
        super(SelfAttentionBlock_APNB, self).__init__()

        self.scale = scale
        self.in_channels = in_channels
        self.out_channels = out_channels
        self.key_channels = key_channels
        self.value_channels = value_channels

        self.pool = nn.Pool2D(pool_size=(scale, scale), pool_type="max")
        self.f_key = layer_libs.ConvBNReLU(
            in_channels=self.in_channels,
            out_channels=self.key_channels,
            kernel_size=1)
        self.f_query = self.f_key
        self.f_value = nn.Conv2d(
            in_channels=self.in_channels,
            out_channels=self.value_channels,
            kernel_size=1)
        self.W = nn.Conv2d(
            in_channels=self.value_channels,
            out_channels=self.out_channels,
            kernel_size=1)

        self.psp_size = psp_size

    def forward(self, x):
        batch_size, _, h, w = x.shape
        if self.scale > 1:
            x = self.pool(x)

        value = self.f_value(x)
        value = _pp_module(value, self.psp_size)
        value = paddle.transpose(value, perm=(0, 2, 1))

        query = self.f_query(x)
        query = paddle.reshape(query, shape=(batch_size, self.key_channels, -1))
        query = paddle.transpose(query, perm=(0, 2, 1))

        key = self.f_key(x)
        key = _pp_module(key, self.psp_size)

        sim_map = paddle.matmul(query, key)
        sim_map = (self.key_channels**-.5) * sim_map
        sim_map = F.softmax(sim_map, axis=-1)

        context = paddle.matmul(sim_map, value)
        context = paddle.transpose(context, perm=(0, 2, 1))
        context = paddle.reshape(
            context, shape=[batch_size, self.value_channels, *x.shape[2:]])
        context = self.W(context)

        return context