yuefanhao/SuperPoint-SuperGlue-TensorRT

Error in SuperPoint building engine

Zsy103 opened this issue · 8 comments

Zsy103 commented 
root@1a88e71715e9:/workspace/SuperPoint-SuperGlue-TensorRT/build# ./superpointglue_image  ../config/config.yaml ../weights/ ${PWD}/../image/image0.png ${PWD}/../image/image1.png
Config file is ../config/config.yaml
model_dir: ../weights/superpoint_v1_sim_int32.onnx
First image size: 320x240
Second image size: 320x240
Building inference engine......
False at plan
Error in SuperPoint building engine. Please check your onnx model path.

The environment used is the provided Docker. But I encountered the above problem. I am certain that the path of the model is correct. I have modified the code of superpoint.cpp as follows:

TensorRTUniquePtr<IHostMemory> plan{builder->buildSerializedNetwork(*network, *config)};
    if (!plan) {
+       std::cout << "False at plan\n";
        return false;
    }

So what should I do to solve this problem？

yuefanhao commented

@Zsy103 I haven't had this kind of problem. What GPU are you using? Let me see if I can reproduce it.

Zsy103 commented

NVIDIA RTX 4090

yuefanhao commented

@Zsy103 There may be a problem with this version of TensorRT working on 4090.

supersupercoconut commented

I encounter the same error using Nvidia 4090

aniatt commented

I've encountered the same issue on NVIDIA 3050

supersupercoconut commented

I've encountered the same issue on NVIDIA 3050

I solved this problem by updating the version of TensorRT

aniatt commented

I'm currently using TensorRT 8.4.1.5. Do you know which TensorRT version has success?

supersupercoconut commented

I'm currently using TensorRT 8.4.1.5. Do you know which TensorRT version has success?

I'm using the TensorRT 8.6 in docker with cuda 11.8 on Nvidia 4090