Linux x86平台推理Paddle-OCRv4出错

[mhanhmk]

• 标题：简洁、精准描述您的问题，例如“最新预测库的 API 文档在哪儿 ”
• 版本、预测库信息：
     1）Paddle Lite 版本：v2.12
     2）Host 环境：
     3）运行设备环境：Ubuntu 24.04
     4）预测后端信息：CPU
• 预测信息
     1）预测 API：C++
     2）预测选项信息：x86、单线程 等
     3）预测库来源：https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.12/inference_lite_lib.x86.linux.with_log.tar.gz
• 复现信息：如为报错，请提供剥离业务环境的可复现预测 demo，用于问题复现
• https://github.com/PaddlePaddle/PaddleOCR/tree/main/deploy/lite
• 问题描述：循环第一次predictor->Run();没问题第二次就出意外了

REC Run----------------------------
I0607 16:57:32.935858  7587 conv_compute.cc:119] invoking directConv
I0607 16:57:32.936663  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.937469  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.937788  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.938114  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.938335  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.941613  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.942692  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.944197  7587 conv_compute.cc:119] invoking directConv
I0607 16:57:32.944640  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.944905  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.947846  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.951077  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.954516  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.958043  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.958745  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.963771  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.969869  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.978529  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.987459  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:32.988895  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:32.997839  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.008172  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.016959  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.029103  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.034193  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.040522  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.049707  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.056203  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:33.057575  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.068104  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.088347  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.094601  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.103175  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.106088  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.111600  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.116544  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.121049  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:33.122073  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.127370  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.136443  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.141999  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.146546  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:33.147318  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.152208  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.161625  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.166610  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.171103  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:33.171985  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.177032  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.186021  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.191164  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.195720  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:33.196465  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.201404  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.210927  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.215935  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.230808  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.233219  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.240857  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.245848  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.250301  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:33.251062  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.256037  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.270426  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.275490  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.289672  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.291671  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.297890  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.299757  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.301578  7587 conv_compute.cc:107] invoking conv_depthwise_3x3p0p1 or conv_depthwise_5x5
I0607 16:57:33.301777  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.303645  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.310194  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
I0607 16:57:33.312065  7587 elementwise_compute.cc:73] Fast broadcast chk fail, for x_dims smaller.
REC Run ----------------------------
double free or corruption (!prev)

[ddchenhao66]

您好～您的问题里说的是在x86 cpu上推理，但是复现环境给的链接是https://github.com/PaddlePaddle/PaddleOCR/tree/main/deploy/lite ，链接里是arm的环境部署步骤。请问您跑的推理代码是什么，是来自于哪个repo?以及跟您确认下模型是否来自于https://github.com/PaddlePaddle/PaddleOCR?tab=readme-ov-file#%EF%B8%8F-pp-ocr-%E7%B3%BB%E5%88%97%E6%A8%A1%E5%9E%8B%E5%88%97%E8%A1%A8%E6%9B%B4%E6%96%B0%E4%B8%AD ？

[mhanhmk]

您好～您的问题里说的是在x86 cpu上推理，但是复现环境给的链接是https://github.com/PaddlePaddle/PaddleOCR/tree/main/deploy/lite ，链接里是arm的环境部署步骤。请问您跑的推理代码是什么，是来自于哪个repo?以及跟您确认下模型是否来自于https://github.com/PaddlePaddle/PaddleOCR?tab=readme-ov-file#%EF%B8%8F-pp-ocr-%E7%B3%BB%E5%88%97%E6%A8%A1%E5%9E%8B%E5%88%97%E8%A1%A8%E6%9B%B4%E6%96%B0%E4%B8%AD ？

复现的链接是arm环境的，我将它移植到了x86平台，这是我的复现连接https://github.com/hanmingk/cxx-ocr/tree/main/paddle_lite_ocr 。模型是从这下的https://github.com/PaddlePaddle/PaddleOCR/blob/main/doc/doc_ch/models_list.md

[ddchenhao66]

如果是要在x86 cpu平台推理，还是建议使用 Paddle Inference https://www.paddlepaddle.org.cn/inference/v2.6/guides/introduction/index_intro.html

[mhanhmk]

如果是要在x86 cpu平台推理，还是建议使用 Paddle Inference https://www.paddlepaddle.org.cn/inference/v2.6/guides/introduction/index_intro.html

我提供的复现链接中也有Paddle Inference平台的推理，它可以正常运行，我想要整合进app里Paddle Inference太大的，不过我后面使用模型转换成功实现了在mindspore lite推理运行。

[ddchenhao66]

如果需要用Paddle Lite平台上跑x86 cpu的推理，可以提供export GLOG_v=5后运行的日志，方便我们这边进一步分析