search

linghu8812 / tensorrt_inference

706 stars 208 forks source link

tensorrt_inference/nanodet/nanodet.cpp:59: bool nanodet::InferenceFolder(const string&): Assertion `engine->getNbBindings() == 2' failed. 已放弃 (核心已转储) #17

Closed Wulingtian closed 3 years ago

Wulingtian commented 3 years ago

当我做nanodet trt推理的时候，报错；我打印了engine->getNbBindings() ，发现engine->getNbBindings() == 7而不是等于2；我检查了网络，发现网络有一个输入和6个输出；请问怎么转换成一个输入和一个输出的模型呢？

KimYJ12 commented 3 years ago

I also encountered the same error. When I remove the part(assert(engine->getNbBindings() == 2)), the following error occurs.

loading filename from:../yolov4-p7.trt deserialize done [12/21/2020-17:28:42] [W] [TRT] Total space of persistent layer space is 486812 on host and 2471966208 on device binding0: 28311552 binding1: 50135040 binding2: 12533760 binding3: 3133440 binding4: 783360 binding5: 195840 Processing: ../samples//eagle.jpg prepareImage prepare image take: 230.207 ms. host2device execute [12/21/2020-17:28:43] [E] [TRT] engine.cpp (1036) - Cuda Error in executeInternal: 700 (an illegal memory access was encountered) [12/21/2020-17:28:43] [E] [TRT] FAILED_EXECUTION: std::exception Inference take: 289.998 ms. execute success device2host post process Segmentation fault (core dumped)

linghu8812 commented 3 years ago

@Wulingtian please use this pro to export nanodet onnx

linghu8812 commented 3 years ago

@KimYJ12 please use this pro to export yolov4-p7 onnx.

AderonHuang commented 3 years ago

当我做nanodet trt推理的时候，报错；我打印了engine->getNbBindings() ，发现engine->getNbBindings() == 7而不是等于2；我检查了网络，发现网络有一个输入和6个输出；请问怎么转换成一个输入和一个输出的模型呢？

楼主请问这个问题解决了吗，我也遇到了这个问题，用的是https://github.com/linghu8812/tensorrt_inference/tree/master/ScaledYOLOv4这个里面的export_onnx.py转换的，和 @linghu8812 提供的是一样的呀