search

NVIDIA / trt-samples-for-hackathon-cn

Simple samples for TensorRT programming
Apache License 2.0
1.5k stars 338 forks source link

关于08-Advance/StreamAndAsync中的示例得到的结果与视频展示不符 #101

Open CabinH opened 10 months ago

CabinH commented 10 months ago

1703229887953_9E27329E-F4FE-4a92-8A6D-937D5DCBCBE0 如图，从模型构建到推理整个流程都是根据08-Advance/StreamAndAsync/main.py中执行

使用参数：

2. Calculation-bound

nB,nC,nH,nW = 8,64,128,128 nCOut,nKernelHeight,nKernelWidth = 64,9,9

与视频中多流的预期结果有很大差距 lQDPJxX5ErUH3TPNBJLNCeSw4Nw2hE4t3LUFdbn7ZkiZAA_2532_1170

实验环境为windows11+TensorRT8.6+3070ti

请问应该如何调整才能实现如视频中多流的效果一致？