使用C++进行端到端多线程推理时：如果把ppocr3模型对象 Clone() 2次，每个线程使用一个unique_ptr<fastdeploy::pipeline::PPOCRv3>预测，程序必出现Mats shapes are not consisten

huangtao2999 commented 9 months ago

环境

[ERROR] fastdeploy/vision/common/processors/mat_batch.cc(33)::Tensor Mats shapes are not consistent.

【模型跑不通】 auto ppocr_v3 = fastdeploy::pipeline::PPOCRv3(&det_model, &cls_model, &rec_model); auto ppocr1 =ppocr_v3.Clone(); auto ppocr2 =ppocr_v3.Clone();

cv::Mat im = imageModel.GetIm(); fastdeploy::vision::OCRResult result; ppocr_v3.Predict(&im, &result);

cv::Mat im1 = imageModel.GetIm(); fastdeploy::vision::OCRResult result1; ppocr1->Predict(&im1, &result1);

cv::Mat im2 = imageModel.GetIm(); fastdeploy::vision::OCRResult result2; ppocr2->Predict(&im2, &result2);

rainyfly commented 7 months ago

已收到反馈，我们分析一下原因

tim7-m commented 4 months ago

已收到反馈，我们分析一下原因

您好，我们再生产环境中遇到了同样的问题，请问这个问题有修复吗？或者这个问题有计划处理吗？谢谢

KyleWang-Hunter commented 4 months ago

真是头疼，这个框架到底能不能在生产环境中用嘛？