SeaLand 分割数据

未分类

1. 环境配置

硬件环境

CPU: Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz
GPU: GeForce GTX 1080 Ti
RAM: 20 GB

1. 得到结果

res-Unet：

框架：MXnet
train：/home/dl/experiment/seaLand/res-try/trainsea.py
size：448*448
batch：11
res:LP=98.58, LR=98.91, OP=99.04, OR=99.04 mean=98.89

初始 0.01
total_steps = 150
step=[total_steps // 2, total_steps // 4 * 3], factor=0.1
实际选择 136th epoch

u-Net:

框架：Keras
train：/home/dl/experiment/seaLand/grss_uNet/uNetForSeaLand.py
size：整图
unet:LP=96.68, LR=97.42, OP=97.57, OR=97.57 mean=97.31

sgd=K.optimizers.SGD(lr=1*1e-3)

SegNet:

框架：caffe
train：/home/dl/leiYang/SegNet/sealandCaffe/segnet_solver.prototxt caffee train -gpu 0 -solver segnet_solver.prototxt &
size：480*360
seg:LP=97.52, LR=96.50, OP=97.81, OR=97.81 mean=97.41

base_lr: 0.001
lr_policy: "step"
momentum: 0.9