硬件配置:GTX 1080,
docker 的镜像配置是CUDA8,,在哪里看到说gtx1080不能用CUDA7.5及以下,todo:找出那个链接
镜像地址为docker pull xiaoxinyi/caffe-gpu:8.23
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~SSD训练
cd caffe之后,git checkout ssd
会出现两个错误:
(1)解决caffe make报错:/usr/include/boost/property_tree/detail/json_parser_read.hpp:257:264: error: ‘type name’ declared as function returning an array:
解决方法:https://blog.csdn.net/allyli0022/article/details/62881238
(2)安装sudo apt-get install libatlas-base-dev(不然编译caffe时候出错/usr/bin/ld:cannot find -lcblas和latlas
(1)预训练模型:链接: https://pan.baidu.com/s/1IzLh-YcW52tseZUldcEEhg 提取码: rprn
(2)修改create_list.sh和create_data的时候,$HOME改成caffe所在路径
(2)在score_ssd_pascal.py/ssd_pascal_webcam.py/ssd_pascal_video.py等对应脚本中添加 sys.path.insert(0,'/home/xxx/caffe/python')
,,,,否则create_data.sh的时候报错放
(3.按3.链接继续修改,运行:python ./examples/ssd/ssd_pascal.py 2>&1 | tee ssd_train_log.txt 开始训练
------------------------------------------------------------------------------------------------
学习率base_lr, batch_size ,max_iteration, snapshot(保存间隔)
save_dir(模型保存路径)
=================================
第二步的报上有人说升级gcc,升级后还是报错gcc version later than 5.3 are not supported;不过按上面的链接修改/usr/include/boost/property_tree/detail/json_parser_read.hpp就可以了