amokame

> 对，哈哈我忘记了，我转成trt之后测速在1080ti上STDC-Seg50能到255帧，STDC-Seg75也有129.8帧，确实效果和速度都是不错的 打扰了，能不能告知一下你用什么转的trt？试了几个都没成功

> > 如果我没记错的话， 按照README.md中的流程，安装pycuda和tensorrt就行，版本可以不跟作者的对应，然后运行python run_latency.py，如果你已经在这个py文件中指定了onnx路径和tensorrt路径，则会先给你导出onnx然后再转成tensorrt就可以了，时间有点久了，细节记得不太清楚了，大概就是这样。核心就是你要保证你的pycuda和tensorrt是安装成功了的 了解了 我去试试。非常感谢！！！

> > > 对，哈哈我忘记了，我转成trt之后测速在1080ti上STDC-Seg50能到255帧，STDC-Seg75也有129.8帧，确实效果和速度都是不错的 > > > > > > 打扰了，能不能告知一下你用什么转的trt？试了几个都没成功 > > 如果我没记错的话， 按照README.md中的流程，安装pycuda和tensorrt就行，版本可以不跟作者的对应，然后运行python run_latency.py，如果你已经在这个py文件中指定了onnx路径和tensorrt路径，则会先给你导出onnx然后再转成tensorrt就可以了，时间有点久了，细节记得不太清楚了，大概就是这样。核心就是你要保证你的pycuda和tensorrt是安装成功了的 不好意思再打扰一下>< 你有没有碰到过下面这个问题？ [TensorRT] ERROR: Network must have at least one output with build_engine("model.onnx") as...

> this helps me out > > [NVIDIA/TensorRT#183 (comment)](https://github.com/NVIDIA/TensorRT/issues/183#issuecomment-657673563) I tried to change the build_engine in latency/utils/darts_utils.py as the following. ``` def build_engine(model_file): explicit_batch = 1

> I had the same issue. I managed to fix this thanks to [this comment](https://github.com/NVIDIA/TensorRT/issues/183#issuecomment-657673563). Sorry to bother you. I tried to change the build_engine in latency/utils/darts_utils.py as the following....

> Maybe changing this parameter ''depth_scale'' would help? How to confirm the value of ''depth_scale''? Sorry, I don't know what this parameter is, also didnot find it in args.py

I tried pre-processed depth and non-pre-processed depth, and it didnot work well for non-pre-processed depth. So at least for my situation, I think pre-processing is necessary.