search

MiniBullLab / easy_docker

1 stars 1 forks source link

easy_docker v0.0.1测试 #6

Closed MiniBullLab closed 4 years ago

MiniBullLab commented 4 years ago

1.分类、检测、分割模型可以训练 2.docker的GPU使用率 3.docker从安装到最终训练出结果，整理流程是否有问题

foww-0001 commented 4 years ago
  1. 分类和分割训练没有问题
  2. 检测训练没有问题，但是转换出现错误： 
    F0920 02:21:40.294873  7322 cudnn_deconv_layer.cpp:53] Check failed: status == CUDNN_STATUS_SUCCESS (1 vs. 0)  CUDNN_STATUS_NOT_INITIALIZED

    @lpj0822

MiniBullLab commented 4 years ago

v0.1以上问题已经解决

MiniBullLab commented 4 years ago

docker 测试相关问题记录在此 @foww-0001

lpj0822 commented 4 years ago

源码在docker中测试

lpj0822 commented 4 years ago

GPU使用率

foww-0001 commented 4 years ago

源码在docker中测试中报错：

Traceback (most recent call last):
  File "easyai/train_task.py", line 5, in <module>
    from easyai.tasks.utility.registry import REGISTERED_TRAIN_TASK
ModuleNotFoundError: No module named 'easyai'

需要把源码的路径写道环境变量中：

export PYTHONPATH=/easy/easy_ai:$PYTHONPATH

修改完后可以正常运行。

源码classnet，detnet，segnet训练无误。 @lpj0822

lpj0822 commented 4 years ago

docker stop $(docker ps -a -q)

lpj0822 commented 4 years ago

docker rm $(docker ps -a -q)

lpj0822 commented 4 years ago

docker rmi $(docker images -q)

foww-0001 commented 4 years ago

最新打包的easyai，easy_converter，easy_tools在docker环境下测试classnet，detnet，segnet结果正常。 @lpj0822

vitahlin commented 4 years ago

v0.0.1测试完成