search

linghu8812 / tensorrt_inference

696 stars 205 forks source link

Some tactics do not have sufficient workspace memory to run. Increasing workspace size may increase performance, please check verbose output. #16

Open bobbilichandu opened 3 years ago

bobbilichandu commented 3 years ago

Facing this issue. Any idea how to solve this?

linghu8812 commented 3 years ago

try modify this line: https://github.com/linghu8812/tensorrt_inference/blob/887cca1487395cc46a23537213201d224600a976/includes/common/common.hpp#L132

bobbilichandu commented 3 years ago

Got it. Is there any way to use this trt file in an api. Calling from a python function or something similar?

lfydegithub commented 3 years ago

try modify this line:

https://github.com/linghu8812/tensorrt_inference/blob/887cca1487395cc46a23537213201d224600a976/includes/common/common.hpp#L132

I am going to use scaled-yolov4-p5, what is the number should i change to? input_shape is 896x896x3, thanks. My computer reboot every time when I run the bin file. GPU is RTX2080Super 8G, CUDA10.2, TensorRT7.1.3.4, torch1.7. It is able to run detect.py.

lfydegithub commented 3 years ago

try modify this line: https://github.com/linghu8812/tensorrt_inference/blob/887cca1487395cc46a23537213201d224600a976/includes/common/common.hpp#L132

I am going to use scaled-yolov4-p5, what is the number should i change to? input_shape is 896x896x3, thanks. My computer reboot every time when I run the bin file. GPU is RTX2080Super 8G, CUDA10.2, TensorRT7.1.3.4, torch1.7. It is able to run detect.py.

I have set to 1000_MiB, case1: scaled-yolov4-p5 + 640x640, it is able to inference. case2: scaled-yolov4-p5 + 896x896, it is still to reboot.

bobbilichandu commented 3 years ago

When you are exporting to onnx, you need to specify the img size. So while using the trt engine, it expects the img size needs to be 640*640.

lfydegithub commented 3 years ago

When you are exporting to onnx, you need to specify the img size. So while using the trt engine, it expects the img size needs to be 640*640.

yeah, I know. ''896x896'' means python export_onnx.py -img-size 896. using the same image shape both export and inference.

bobbilichandu commented 3 years ago

Did you edit the config file while exporting?