PaddleOCR icon indicating copy to clipboard operation
PaddleOCR copied to clipboard

Paddle Serving推理部署QPS比官方数据低非常多

Open pooregg opened this issue 2 years ago • 0 comments

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem

  • 系统环境/System Environment:
  • 版本号/Version:2.4 Paddle: PaddleOCR: 问题相关组件/Related components:
  • 运行指令/Command Code:
  • 完整报错/Complete Error Message: 2022-07-28 08:33:01,043 Op(rec): 2022-07-28 08:33:01,043 in[1347.551125 ms] 2022-07-28 08:33:01,044 prep[62.27275 ms] 2022-07-28 08:33:01,044 midp[796.63025 ms] 2022-07-28 08:33:01,044 postp[94.265125 ms] 2022-07-28 08:33:01,044 out[1.04975 ms] 2022-07-28 08:33:01,044 idle[0.585897574865245] 2022-07-28 08:33:01,044 Op(det): 2022-07-28 08:33:01,044 in[7057.73925 ms] 2022-07-28 08:33:01,044 prep[106.721625 ms] 2022-07-28 08:33:01,045 midp[37.337875 ms] 2022-07-28 08:33:01,045 postp[58.09525 ms] 2022-07-28 08:33:01,045 out[2.02075 ms] 2022-07-28 08:33:01,045 idle[0.9721623350095097] 2022-07-28 08:33:01,045 DAGExecutor: 2022-07-28 08:33:01,045 Query count[8] 2022-07-28 08:33:01,045 QPS[0.8 q/s] 2022-07-28 08:33:01,045 Succ[1.0] 2022-07-28 08:33:01,046 Error req[]

对比了https://aistudio.baidu.com/aistudio/projectdetail/3630726 部署教程中的pipeline.tracer文件中的数据,发现我Op(rec)-in Op(det)-in 的耗时都非常大,不知道是什么原因。图片大小是1920x1080的

pooregg avatar Aug 05 '22 07:08 pooregg