PaddleX模型faster rcnn 训练报错（batch Size）设置过大

yellowopen.hk

于 2021-03-26 16:18:39 发布

阅读量801

点赞数

分类专栏： PaddleX试错公告栏文章标签：深度学习

本文链接：https://blog.youkuaiyun.com/firstpmhk/article/details/115249253

版权

在使用PaddleX进行工业质检铝表面检测的深度学习任务时，若不采用DCN，遇到训练报错提示批大小(batch Size)设置过大。为解决此问题，建议降低batch Size的设定。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

2021-03-26 07:03:48 [INFO]      Starting to read file list from dataset...
2021-03-26 07:03:58 [INFO]      5316 samples in file ./alu/aluminum_inspection/train_list.txt
creating index...
index created!
2021-03-26 07:03:58 [INFO]      Starting to read file list from dataset...
2021-03-26 07:03:58 [INFO]      285 samples in file ./alu/aluminum_inspection/val_list.txt
creating index...
index created!
W0326 07:04:57.350690 34729 device_context.cc:252] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 11.0, Runtime API Version: 10.0
W0326 07:04:57.353631 34729 device_context.cc:260] device: 0, cuDNN Version: 8.0.
W0326 07:05:02.598165 34729 fuse_all_reduce_op_pass.cc:74] Find all_reduce operators: 72. To make the speed faster, some all_reduce ops are fused during training, after fusion, the number of all_reduce ops is 72.
W0326 07:05:04.521656 34929 init.cc:226] Warning: PaddlePaddle catches a failure signal, it may not work properly
W0326 07:05:04.521698 34929 ini

最低0.47元/天解锁文章