Pytorch 深度学习特征图可视化——（Yolo网络、rcnn）
写在前面
想系统学习深度学西可查看weibuweibu123此人博文
下面代码可以对实现特征图的可视化；主要以Yolov4作为例子，使用者仅仅修改自己想可视乎网络的层数或者某层的任意维度
目录
写在前面
对Yolo系列、r-cnn系列都有较好的表现形式，下面我们以Yolov4网络做详细介绍
1：获得网络的模型结构：
Yolobody(
(backbone): CSPDarkNet(
(conv1): BasicConv(
(conv): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(mish): Mish()
)
(stages): ModuleList(
(0): Resblock_body(
(downsaple_conv): BasicConv(
(conv): Conv2d(32, 64, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(mish): Mish()
)
(split_conv0): BasicConv(
(conv): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(mish): Mish()
)
(split_conv1): BasicConv(
(conv): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(mish): Mish()
)
(block_conv): Sequential(
(0): Resblock(
(block): Sequential(
(0): BasicConv(
(conv): Conv2d(64, 32, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(mish): Mish()
)
(1): BasicConv(
(conv): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(mish): Mish()
)
)
)2：定义钩子函数
# 定义钩子函数，获取指定层名称的特征
activation = {} # 保存获取的输出
def get_activation(name):
def hook(model, input, output):
activation[name] = output.detach()
return hook
3：根据钩子函数可视化特征图
比如我们想可视化 上述模型中倒数第六行的卷积层的特征图，我们从模型上发现他的模型位置是：backbone->stage[0]->block_conv[0]->block[1]->conv !!!!看（）里面的
4：详细代码
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
from torch.nn import functional as F
from torchvision import transforms
import numpy as np
from PIL import Image
from collections import OrderedDict
import cv2
from net.yolov4 import Yolobody
# 定义钩子函数，获取指定层名称的特征
activation = {} # 保存获取的输出
def get_activation(name):
def hook(model, input, output):
activation[name] = output.detach()
return hook
def main():
anchors_mask = [[6, 7, 8], [3, 4, 5], [0, 1, 2]]
#调用模型
model = Yolobody(anchors_mask,20,False)
# print(model)
model.eval()
# 根据刚刚指定的层数更改
model.backbone.stages[0].block_conv[0].block[1].conv.register_forward_hook(get_activation('conv'))
_,_,_ = model(input_img)
bn = activation['conv'] # 结果将保存在activation字典中
print(bn.shape)
#显示特征图，通过更改下面的代码可以看到64维中任何几维
plt.figure(figsize=(12,12))
for i in range(64):
if i <63:
continue
else:
j=i
j = j-63
plt.subplot(1,2,j+1)
#显示每个通道的特征图因此用灰色显示
plt.imshow(bn[0,i,:,:], cmap='gray')
plt.axis('off')
plt.show()
return 0
if __name__=='__main__':
transform = transforms.Compose([
transforms.Resize([416,416]),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
# 从测试集中读取一张图片，并显示出来
img_path = 'F:/tupian/0048.jpg'
img = Image.open(img_path)
#给图片增加一个N维度
input_img = transform(img).unsqueeze(0)
#[1,3,224,224]
print(input_img.shape)
# imgarray = np.array(img) / 255.0
# plt.figure(figsize=(8,8))
# plt.imshow(imgarray)
# plt.axis('off')
# plt.title("the first picture")
# plt.show()
# #将图片resize。精度变低
# image = img.resize([416,416])
# imgarray = np.array(image) / 255.0
# plt.figure(figsize=(8,8))
# plt.imshow(imgarray)
# plt.title("the second picture")
# plt.show()
main()
# 将图片处理成模型可以预测的形式
5：可视化结果
