Fully Convolutional Networks for Semantic Segmentation
——————————————————————————————————————————
【FCN实践】01 常见问题 http://blog.csdn.net/binlearning/article/details/72854136
【FCN实践】02 模型迁移及初始化 http://blog.csdn.net/binlearning/article/details/72854244
【FCN实践】03 训练 http://blog.csdn.net/binlearning/article/details/72854407
【FCN实践】04 预测 http://blog.csdn.net/binlearning/article/details/72854583
【项目源码】https://github.com/binLearning/fcn_voc_32s
——————————————————————————————————————————
经过参数迁移及初始化以后,可以用两种方法来进行训练。一是caffe命令行形式,二是python程序形式。
1.命令行形式
path-to-caffe/build/tools/caffe train -solver solver.prototxt -weights voc_fcn32s_trans_init.caffemodel -gpu 0
solver. prototxt
train_net: "./model/voc_fcn32s_train.prototxt"
test_net: "./model/voc_fcn32s_val.prototxt"
test_iter: 736
# make test net, but don't invoke it from the solver itself
test_interval: 8498
display: 100
average_loss: 20
lr_policy: "fixed"
# lr for unnormalized softmax
base_lr: 1e-10
# high momentum
momentum: 0.99
# no gradient accumulation
iter_size: 1
max_iter: 849800
weight_decay: 0.0005
snapshot: 8498
snapshot_prefix: "snapshot/voc_fcn32s"
test_initialization: false
注:
①模型需要先经过参数迁移及初始化处理;
②solver. prototxt中需要设置有效的测试间隔(test_interval);
③python layer操作实现文件(voc_layers.py)放到path-to-caffe/python文件夹中。
2.python程序形式
根据官网开源程序修改。
import os
import sys
import numpy as np
import caffe
import score
weights = './model/voc_fcn32s_trans_init.caffemodel'
# init
caffe.set_device(0)
caffe.set_mode_gpu()
solver = caffe.SGDSolver('solver.prototxt')
solver.net.copy_from(weights)
# scoring
val = np.loadtxt('./data/segvalid11.txt', dtype=str)
for _ in range(100):
solver.step(8498)
#score.seg_tests(solver, False, val, layer='score')
score.seg_tests(solver, 'val_save_{0}', val, layer='score')
注:
①测试间隔由solver.step(8498)中的数值控制;
②score.seg_tests函数中的第二个形参如果有效则会存储测试时的score图像。
3.改进score图像
原程序中保存图像为灰度图,标签值即像素值,但标签值很小([1,20])所以预测结果不容易观察,现改成以对应的颜色来表示物体,保存为颜色图。
4.训练概况
5.训练各阶段预测示例
2007_000033
2007_000491
2007_009521