实验记录04.22.2023
sixwalter Lv6

实验记录04.22.2023

实验一:处理joints数据

  • 生成注释文件,便于统一读取:
1
find /mnt/data1/data/volleyball/joints/*/*.pickle | sed -E 's/(\/mnt\/data1\/data\/volleyball\/joints\/([0-9]*)\/([0-9]*).pickle)/\1 \2 \3/' > data/volleyball/joints/ann.txt
image-20230422165338524
  • 读取注释文件:
image-20230422170312052

这种方式可以把数据统一读取到内存里,不需要get数据的时候从pickle文件里读,所以会训练会快一点

实验二:修改网络结构

  • 对特征进行预处理,并加入pose_head模块:
image-20230424124744155
  • 损失函数计算和指标计算:
image-20230424124957978
  • 实验结果:
image-20230424125151924
  • Post title:实验记录04.22.2023
  • Post author:sixwalter
  • Create time:2023-04-22 00:00:00
  • Post link:https://coelien.github.io/2023/04/22/deep-learning/第一章/实验记录4.22/
  • Copyright Notice:All articles in this blog are licensed under BY-NC-SA unless stating additionally.
 Comments