115 lines
3.9 KiB
Python
115 lines
3.9 KiB
Python
#!/usr/bin/env python
|
||
# -*- coding: utf-8 -*-
|
||
"""
|
||
@project:
|
||
@File : parts
|
||
@Author : qiqq
|
||
@create_time : 2023/5/30 14:18
|
||
"""
|
||
|
||
import numpy as np
|
||
import torch
|
||
import torch.nn.functional as F
|
||
from torch import nn
|
||
|
||
|
||
|
||
class SpatialGather_Module(nn.Module):
|
||
"""
|
||
|
||
Aggregate the context features according to the initial
|
||
predicted probability distribution.
|
||
Employ the soft-weighted method to aggregate the context.
|
||
"""
|
||
|
||
def __init__(self, cls_num=2, scale=1):
|
||
super(SpatialGather_Module, self).__init__()
|
||
self.cls_num = cls_num
|
||
self.scale = scale
|
||
|
||
def forward(self, feats, probs):
|
||
# 以原始输入512为例 feats 512*64*64 probs num*64*64
|
||
batch_size, c, h, w = probs.size(0), probs.size(1), probs.size(2), probs.size(3)
|
||
probs = probs.view(batch_size, c, -1) # 2*1024
|
||
feats = feats.view(batch_size, feats.size(1), -1) # 512*1024
|
||
feats = feats.permute(0, 2, 1) # batch x hw x c #1024*512
|
||
probs = F.softmax(self.scale * probs, dim=2) # batch x k x hw #2*1024 #把它转成概率
|
||
ocr_context = torch.matmul(probs, feats) \
|
||
.permute(0, 2, 1).unsqueeze(3) # numclass，1024 * 1024*512--》2*512（这就是那个类别特征）---》batch 512 2 1
|
||
|
||
pv_context=ocr_context[:,:,1,:].unsqueeze(dim=-1) #batch channel h w
|
||
return pv_context # 维度是batch 512 numclass 1
|
||
|
||
# class SFAttention(nn.Module):
|
||
# def __init__(self,feats_channels=256,numclass=2):
|
||
# super(SFAttention, self).__init__()
|
||
# self.project1 = nn.Sequential(
|
||
# nn.Conv2d(256, feats_channels, 1, bias=False),
|
||
# nn.BatchNorm2d(feats_channels),
|
||
# nn.ReLU(inplace=True))
|
||
# self.auxproject = nn.Conv2d(feats_channels, numclass, 1, bias=False)
|
||
# self.getpvf=SpatialGather_Module()
|
||
# self.normalizer = nn.Sigmoid()
|
||
#
|
||
#
|
||
# def forward(self,x,):
|
||
# '''思路：
|
||
# 先弄出一个代表光伏的向量，
|
||
# 然后这个向量和原始的特征图的每个点做内积然后经过sigmode形成一个注意力图
|
||
#
|
||
# 这个注意力图与原特征再相乘
|
||
# 关于这个pv的向量怎来：
|
||
# 参考ocr
|
||
# '''
|
||
# inputs = self.project1(x) # --256
|
||
# preds = self.auxproject(inputs) #这个玩意不一定由这个input生成可以由别的
|
||
# pv_feature=self.getpvf(inputs,preds)
|
||
#
|
||
# #
|
||
# relations = self.normalizer((pv_feature * inputs).sum(dim=1, keepdim=True)) #每个像素点和pvpv_feature算一个相似度，然后形成一个o-1的权重
|
||
#
|
||
# refined_feats = relations * inputs #相当于给每个像素点加权
|
||
#
|
||
# return refined_feats,preds
|
||
|
||
|
||
|
||
|
||
|
||
# model = SFAttention()
|
||
# #
|
||
# inpoo=torch.rand(8,256,64,64)
|
||
# # pre=torch.rand(8,2,64,64)
|
||
# out=model(inpoo)
|
||
# print(out.shape)
|
||
|
||
class SFAttention(nn.Module):
|
||
def __init__(self,feats_channels=256,numclass=2):
|
||
super(SFAttention, self).__init__()
|
||
self.project1 = nn.Sequential(
|
||
nn.Conv2d(256, feats_channels, 1, bias=False),
|
||
nn.BatchNorm2d(feats_channels),
|
||
nn.ReLU(inplace=True))
|
||
|
||
self.normalizer = nn.Sigmoid()
|
||
|
||
|
||
def forward(self,x,pv_feature):
|
||
'''思路：
|
||
先弄出一个代表光伏的向量，
|
||
然后这个向量和原始的特征图的每个点做内积然后经过sigmode形成一个注意力图
|
||
|
||
这个注意力图与原特征再相乘
|
||
关于这个pv的向量怎来：
|
||
参考ocr
|
||
'''
|
||
inputs = self.project1(x) # --256
|
||
|
||
#
|
||
relations = self.normalizer((pv_feature * inputs).sum(dim=1, keepdim=True)) #每个像素点和pvpv_feature算一个相似度，然后形成一个o-1的权重
|
||
|
||
refined_feats = relations * inputs #相当于给每个像素点加权
|
||
return refined_feats
|
||
|
||
|