TRT8版 crash

[tuteming]

我用master版by TensorRT-7.2.1.6, 一切ok.

當用TRT8版by TensorRT-8.0.1.6, compile ok, yolo-trt.exe and detector.dll 也都產生 run yolo-trt.exe,出現 File does not exist : ../configs/yolov4-kHALF-batch1.engine Loading pre-trained weights... Loading complete! ........ 過一下yolo-trt.ex就跳掉了 請賜教,謝謝

[enazoe]

@tuteming 看下trt和cuda cudnn的版本是不是匹配，这种问题一般情况下是环境问题

[tuteming]

trt和cuda cudnn的版本是是匹配的

[tuteming]

另一個問題,我用master版跑 偵測都對,but in command 視窗,出現的是無窮迴圈,同一張圖一直偵測,輸出訊息一直跑, 請問這是甚麼問題?也請賜教,謝謝 Loading TRT Engine... Loading Complete! pre elasped time:3.3799ms inference elasped time:318.334ms post elasped time:0.4608ms detect elasped time:324.954ms batch 0 id:4 prob:0.999942 rect:[114 x 131 from (621, 611)] batch 0 id:4 prob:0.999902 rect:[209 x 63 from (199, 488)] batch 0 id:4 prob:0.999899 rect:[250 x 81 from (185, 415)] batch 0 id:4 prob:0.999696 rect:[210 x 67 from (211, 455)] batch 0 id:4 prob:0.999689 rect:[32 x 106 from (97, 434)] batch 0 id:4 prob:0.998361 rect:[109 x 137 from (605, 643)] pre elasped time:3.3904ms inference elasped time:20.962ms post elasped time:0.4103ms detect elasped time:26.4252ms batch 0 id:4 prob:0.999942 rect:[114 x 131 from (621, 611)] batch 0 id:4 prob:0.999902 rect:[209 x 63 from (199, 488)] batch 0 id:4 prob:0.999899 rect:[250 x 81 from (185, 415)] batch 0 id:4 prob:0.999696 rect:[210 x 67 from (211, 455)] batch 0 id:4 prob:0.999689 rect:[32 x 106 from (97, 434)] batch 0 id:4 prob:0.998361 rect:[109 x 137 from (605, 643)] pre elasped time:4.7177ms inference elasped time:21.1883ms post elasped time:0.4125ms detect elasped time:28.917ms .......

[enazoe]

程序里有for循环，看一眼代码就知道了

[tuteming]

請問你指的for循环是哪裡?下面我mark loop,結果還是一樣. int i = 0;// only one image //disp //for (int i = 0; i < batch_img.size(); ++i) //{ for (const auto &r : batch_res[i]) { std::cout << "batch " << i << " id:" << r.id << " prob:" << r.prob << " rect:" << r.rect << std::endl; cv::rectangle(batch_img[i], r.rect, cv::Scalar(255, 0, 0), 2); std::stringstream stream; stream << std::fixed << std::setprecision(2) << "id:" << r.id << " score:" << r.prob; cv::putText(batch_img[i], stream.str(), cv::Point(r.rect.x, r.rect.y - 5), 0, 0.5, cv::Scalar(0, 0, 255), 2); } cv::namedWindow("image" + std::to_string(i), cv::WINDOW_NORMAL); cv::imshow("image" + std::to_string(i), batch_img[i]); //}

[tuteming]

thanks, I known.