FCN tensorflow复现
0. 下载及参考链接
tensorflow fcn 原址
更改参考
权重下载
数据集下载
1. 复现步骤
1.1 电脑配置
win10
anaconda
tensorflow CPU（8个）
1.2 具体执行步骤
- 代码下载更改参考
- 数据集下载完后放置在"FCN.tensorflow\Data_zoo"路径下
- 权重下载完后放置在"\FCN.tensorflow\Model_zoo"路径下
- 管理员身份进入AnacondaPrompt，执行
conda activate py35
cd /d "D:\spyder_project\FCN.tensorflow"
python FCN.py
- 复现结果
(py35) D:\my_project\spyder_project\FCN.tensorflow>python FCN.py
setting up vgg initialized conv layers ...
原始图像： (?, 224, 224, 3)
mean: [[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
...
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]]
获取图片像素均值mean_pixel: [123.68 116.779 103.939]
预处理后的图像: (?, 224, 224, 3)
开始建立VGG网络：
当前形状： (?, 224, 224, 64)
当前形状： (?, 224, 224, 64)
当前形状： (?, 112, 112, 64)
当前形状： (?, 112, 112, 128)
当前形状： (?, 112, 112, 128)
当前形状： (?, 56, 56, 128)
当前形状： (?, 56, 56, 256)
当前形状： (?, 56, 56, 256)
当前形状： (?, 56, 56, 256)
当前形状： (?, 56, 56, 256)
当前形状： (?, 28, 28, 256)
当前形状： (?, 28, 28, 512)
当前形状： (?, 28, 28, 512)
当前形状： (?, 28, 28, 512)
当前形状： (?, 28, 28, 512)
当前形状： (?, 14, 14, 512)
当前形状： (?, 14, 14, 512)
当前形状： (?, 14, 14, 512)
当前形状： (?, 14, 14, 512)
当前形状： (?, 14, 14, 512)
VGG处理后的图像： (?, 14, 14, 512)
pool5： (?, 7, 7, 512)
conv6: (?, 7, 7, 4096)
conv7: (?, 7, 7, 4096)
conv8: (?, 7, 7, 151)
pool4 and de_conv8 ==> fuse1: (?, 14, 14, 512)
pool3 and deconv_fuse1 ==> fuse2: (?, 28, 28, 256)
conv_t3: [Dimension(224), Dimension(224), 151]
Setting up image reader...
No. of training files: 20210
No. of validation files: 2000
Pickling ...
训练集的大小: 20210
验证集的大小: 2000
Setting up dataset reader
Initializing Batch Dataset Reader...
{'resize_size': 224, 'resize': True}
(20210, 224, 224, 3)
(20210, 224, 224, 1)
Initializing Batch Dataset Reader...
{'resize_size': 224, 'resize': True}
(2000, 224, 224, 3)
(2000, 224, 224, 1)
Setting up Saver...
(2, 224, 224, 3) (2, 224, 224, 1)
step: 0
Step: 0, Train_loss:458.508
2019-03-11 10:03:07.694517 ---> Validation_loss: 395.583
(2, 224, 224, 3) (2, 224, 224, 1)
step: 1
(2, 224, 224, 3) (2, 224, 224, 1)
step: 2
(2, 224, 224, 3) (2, 224, 224, 1)
step: 3
(2, 224, 224, 3) (2, 224, 224, 1)
step: 4
(2, 224, 224, 3) (2, 224, 224, 1)
step: 5
(2, 224, 224, 3) (2, 224, 224, 1)
step: 6
(2, 224, 224, 3) (2, 224, 224, 1)
step: 7
(2, 224, 224, 3) (2, 224, 224, 1)
step: 8
(2, 224, 224, 3) (2, 224, 224, 1)
step: 9
(2, 224, 224, 3) (2, 224, 224, 1)
step: 10
Step: 10, Train_loss:33.9817
(2, 224, 224, 3) (2, 224, 224, 1)
step: 11
(2, 224, 224, 3) (2, 224, 224, 1)
step: 12
(2, 224, 224, 3) (2, 224, 224, 1)
step: 13
(2, 224, 224, 3) (2, 224, 224, 1)
step: 14
(2, 224, 224, 3) (2, 224, 224, 1)
step: 15
(2, 224, 224, 3) (2, 224, 224, 1)
step: 16
(2, 224, 224, 3) (2, 224, 224, 1)
step: 17
(2, 224, 224, 3) (2, 224, 224, 1)
step: 18
(2, 224, 224, 3) (2, 224, 224, 1)
step: 19
(2, 224, 224, 3) (2, 224, 224, 1)
step: 20
Step: 20, Train_loss:9.79859
(2, 224, 224, 3) (2, 224, 224, 1)
step: 21
(2, 224, 224, 3) (2, 224, 224, 1)
step: 22
(2, 224, 224, 3) (2, 224, 224, 1)
step: 23
(2, 224, 224, 3) (2, 224, 224, 1)
step: 24
(2, 224, 224, 3) (2, 224, 224, 1)
step: 25
(2, 224, 224, 3) (2, 224, 224, 1)
step: 26
(2, 224, 224, 3) (2, 224, 224, 1)
step: 27
(2, 224, 224, 3) (2, 224, 224, 1)
step: 28
(2, 224, 224, 3) (2, 224, 224, 1)
step: 29
(2, 224, 224, 3) (2, 224, 224, 1)
step: 30
Step: 30, Train_loss:5.26892
(2, 224, 224, 3) (2, 224, 224, 1)
step: 31
(2, 224, 224, 3) (2, 224, 224, 1)
step: 32
(2, 224, 224, 3) (2, 224, 224, 1)
step: 33
(2, 224, 224, 3) (2, 224, 224, 1)
step: 34
(2, 224, 224, 3) (2, 224, 224, 1)
step: 35
(2, 224, 224, 3) (2, 224, 224, 1)
- 查看CPU使用情况
总共有8个CPU，全部利用起来了 - visualize
(py35) D:\my_project\spyder_project\FCN.tensorflow>python FCN.py
setting up vgg initialized conv layers ...
原始图像： (?, 224, 224, 3)
mean: [[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
...
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]
[[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]
...
[123.68 116.779 103.939]
[123.68 116.779 103.939]
[123.68 116.779 103.939]]]
获取图片像素均值mean_pixel: [123.68 116.779 103.939]
预处理后的图像: (?, 224, 224, 3)
开始建立VGG网络：
当前形状： (?, 224, 224, 64)
当前形状： (?, 224, 224, 64)
当前形状： (?, 112, 112, 64)
当前形状： (?, 112, 112, 128)
当前形状： (?, 112, 112, 128)
当前形状： (?, 56, 56, 128)
当前形状： (?, 56, 56, 256)
当前形状： (?, 56, 56, 256)
当前形状： (?, 56, 56, 256)
当前形状： (?, 56, 56, 256)
当前形状： (?, 28, 28, 256)
当前形状： (?, 28, 28, 512)
当前形状： (?, 28, 28, 512)
当前形状： (?, 28, 28, 512)
当前形状： (?, 28, 28, 512)
当前形状： (?, 14, 14, 512)
当前形状： (?, 14, 14, 512)
当前形状： (?, 14, 14, 512)
当前形状： (?, 14, 14, 512)
当前形状： (?, 14, 14, 512)
VGG处理后的图像： (?, 14, 14, 512)
pool5： (?, 7, 7, 512)
conv6: (?, 7, 7, 4096)
conv7: (?, 7, 7, 4096)
conv8: (?, 7, 7, 151)
pool4 and de_conv8 ==> fuse1: (?, 14, 14, 512)
pool3 and deconv_fuse1 ==> fuse2: (?, 28, 28, 256)
conv_t3: [Dimension(224), Dimension(224), 151]
Setting up image reader...
Found pickle file!
训练集的大小: 20210
验证集的大小: 2000
Setting up dataset reader
Initializing Batch Dataset Reader...
{'resize': True, 'resize_size': 224}
(2000, 224, 224, 3)
(2000, 224, 224, 1)
Setting up Saver...
Model restored...
Saved image: 0
Saved image: 1
- test