rcnn on windows #913

sctrueew · 2022-02-16T09:38:13Z

Hi,

I have a problem with the building of rcnn (detectron2) on Windows. I get an error:
Error LNK1181 cannot open input file 'Release\myplugins.lib' rcnn \tensorrtx\rcnn\b\LINK 1
Myplugins.lib isn't in the path, but it built successfully

OpenCV: 4.5.4
TensorRT: TensorRT-7.2.3.4
CUDA: 11.2

The text was updated successfully, but these errors were encountered:

rgkannan676 · 2022-03-10T09:25:11Z

@sctrueew , Hi any update on this issue? I am also facing same issue.

LNK1181 cannot open input file 'Release\myplugins.lib' rcnn
LNK1181 cannot open input file 'C:\Users\tensorrtx\rcnn\build\CMakeFiles\myplugins.dir\Release\myplugins_generated_BatchedNms.cu.obj' myplugins

Thanks

rgkannan676 · 2022-03-10T09:30:06Z

@freedenS . Hi can you please tell what are the steps to make and build RCNN in windows. I am trying to make using Cmake and build using visual studio 2019.

rgkannan676 · 2022-03-10T09:30:48Z

I made the changes - #521 (comment) for cuda 11.1.

rgkannan676 · 2022-03-10T10:01:15Z

Hi, Seems like most of the issues are associated with C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\include\cub\device\dispatch\dispatch_radix_sort.cuh . Some of the issues are as below.

E0276	name followed by '::' must be a class or namespace name	myplugins	C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\include\cub\device\dispatch\dispatch_radix_sort.cuh	410	
E0020	identifier "RADIX_RANK_MATCH" is undefined	myplugins	C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\include\cub\device\dispatch\dispatch_radix_sort.cuh	701

Do I need to add any additional libraries? Please advise.

rgkannan676 · 2022-03-10T10:17:12Z

E0864 DeviceRadixSortPolicy is not a template myplugins C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\include\cub\device\dispatch\dispatch_radix_sort.cuh 824

sctrueew · 2022-03-10T10:41:43Z

@rgkannan676 Hi,

I couldn't build it. I got this error
invalid numeric argument '/Wfatal-errors'

TensorRt: 7.2
Cuda: 10.2
Os: Win 10

rgkannan676 · 2022-03-10T11:02:59Z

Hi @freedenS , Is it possible to build in windows? Can you please confirm? Thank you.

freedenS · 2022-03-11T02:12:00Z

to build rcnn in windows:

open a project with cuda in vs
add rcnn .h .hpp .cpp .cu
set trt/opencv include and directory
add "--extended-lambda" in CUDA C/C++ -> Command Line -> Other options
replace linux function in dirent.h with corresponding win function

freedenS · 2022-03-11T02:15:02Z

@rgkannan676 Hi,

I couldn't build it. I got this error invalid numeric argument '/Wfatal-errors'

TensorRt: 7.2 Cuda: 10.2 Os: Win 10

if you use cuda 10.2, you should assure it is updated. NVIDIA/TensorRT#1151

rgkannan676 · 2022-03-11T03:17:22Z

to build rcnn in windows:

open a project with cuda in vs

add rcnn .h .hpp .cpp .cu

set trt/opencv include and directory

add "--extended-lambda" in CUDA C/C++ -> Command Line -> Other options

replace linux function in dirent.h with corresponding win function

Hi @freedenS , Thank you for the reply. This resolved my issues and was able to build the .exe file.

sctrueew · 2022-03-11T07:59:17Z

@rgkannan676 Hi,

I created a empty project and all .hpp .cpp .cu

TensorRT-7.2.3.4.Windows10.x86_64.cuda-11.0.cudnn8.1
opencv_4_5_4_cuda_11_2
CUDA v11.2
VS 2019

I added --extended-lambda to the Additional options but I get this errors

Could you please some help?

rgkannan676 · 2022-03-11T08:02:59Z

Try to add the libraries.

Linker -> All Options -> Additional Dependencies. Add the following libraries.

cudart_static.lib
nvinfer.lib
cudart.lib
nvinfer_plugin.lib
cudadevrt.lib
cublas.lib
cublasLt.lib

rgkannan676 · 2022-03-11T08:04:56Z

Have you converted .cu files Item Type as "CUDA C/C++" ? You can see this option in the "Properties" of the files.

rgkannan676 · 2022-03-11T08:06:24Z

Also have you added CUDA 11.2 as dependency? Build Dependencies -> Build Customizations and select CUDA 11.2

rgkannan676 · 2022-03-11T08:12:40Z

once you add CUDA 11.2 dependency, "--extended-lambda " need to be added to CUDA C/C++ -> Command Line -> Additional Options.

sctrueew · 2022-03-11T08:14:57Z

I set

and have

opencv_world454.lib
nvinfer.lib
cudart.lib
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\lib\x64\cudart.lib
kernel32.lib
user32.lib
gdi32.lib
winspool.lib
shell32.lib
ole32.lib
oleaut32.lib
uuid.lib
comdlg32.lib
advapi32.lib
cudart_static.lib
nvinfer_plugin.lib
cudadevrt.lib
cublas.lib
cublasLt.lib

but it wan't solved

sctrueew · 2022-03-11T08:33:33Z

@rgkannan676 Have you any idea?

rgkannan676 · 2022-03-11T08:41:14Z

Have you set

C/C++ -> All Options -> Additional Include Directories with following ? (Change path according to your env). You can add the tensorRT include path also. I am not adding since I copied it to CUDA folder.


C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\include
C:\OpenCV\include

Linker -> All Options -> Additional Library directories. Add following. You can add your tensorRT lib path also.

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\lib\x64

Build the project in Release configuration.

sctrueew · 2022-03-11T08:49:46Z

Yes exactly, I set

and

also I have changed #521 (comment)

rgkannan676 · 2022-03-11T09:02:49Z

above are the changes I made. One more thing, Have you checked 'CUDA_PATH' in Environment Variables? Is it pointing to 11.2? If not, set it to 11.2 and restart.

sctrueew · 2022-03-11T09:15:03Z

above are the changes I made. One more thing, Have you checked 'CUDA_PATH' in Environment Variables? Is it pointing to 11.2? If not, set it to 11.2 and restart.

I've set everything up, but I don't know what the problem is?
Could you please share your project or send it to me?

sctrueew · 2022-03-11T09:54:53Z

@rgkannan676
With the configs, I almost built all the projects in this repo except rcnn. Can you tell me what's wrong?

freedenS · 2022-03-11T10:00:11Z

could you show your compiling log?

sctrueew · 2022-03-11T10:02:50Z

could you show your compiling log?

@freedenS Hi,

log.txt

sctrueew · 2022-03-11T10:16:50Z

could you show your compiling log?

@freedenS Hi,

I've added the log.

rgkannan676 · 2022-03-11T10:36:18Z

Hi @sctrueew , Sorry, not able to share project due to privacy issue. One more check, have you set the Item Type property?

sctrueew · 2022-03-11T10:43:50Z

Hi @sctrueew , Sorry, not able to share project due to privacy issue. One more check, have you set the Item Type property?

No problem, yes I checked again and the Item type was set for all *.cu files

freedenS · 2022-03-11T10:49:11Z

1>D:\Projects\yolov5\tensorrtx\rcnn\MaskRcnnInferencePlugin.h(37): error : expected a ";"
it seems like you are not same with master

sctrueew · 2022-03-11T10:53:49Z

1>D:\Projects\yolov5\tensorrtx\rcnn\MaskRcnnInferencePlugin.h(37): error : expected a ";" it seems like you are not same with master

I've cloned again and built it.

rgkannan676 · 2022-03-11T11:01:55Z

FYI, I created a new solution -> new project-> copied all .cpp, .h, .cu files in the git repo to new project folder and added it to project. Replaced the contents in the main function with rcnn.cpp, along with the changes mentioned above.

sctrueew · 2022-03-11T11:23:06Z

FYI, I created a new solution -> new project-> copied all .cpp, .h, .cu files in the git repo to new project folder and added it to project. Replaced the contents in the main function with rcnn.cpp, along with the changes mentioned above.

Once again, I followed the steps step by step.

freedenS · 2022-03-11T11:23:21Z

you'd better use trt7, such as 7.2.1 described in README, it looks like trt8 has changed that api. i'm not sure.

sctrueew · 2022-03-11T11:33:25Z

@freedenS @rgkannan676 Thank you for the tips, finally, I've built it with Tr 7.2.3

sctrueew · 2022-03-11T14:53:14Z

@freedenS Hi,

I've trained a model with mask_rcnn_R_50_FPN_3x and convert the .pth to .wts. I have 1 class and I changed NUM_CLASSES = 1 in gen_wts.py. I could generate the faster.wts successfully, I change in rcnn side lines:

NUM_CLASSES = 1
BACKBONE_RESNETTYPE R50
STRIDE_IN_1X1 true

but I'm facing this error when I run rcnn -s faster.wts faster.engine m:

Loading weights: faster.wts [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 Building engine, please wait for a while... [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: kernel weights has count 0 but 9408 was expected [03/11/2022-18:22:33] [E] [TRT] (Unnamed Layer* 6) [Convolution]: count of 0 weights in kernel, but kernel dimensions (7,7) with 3 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 3 * 7*7 * 64 / 1 = 9408 [03/11/2022-18:22:33] [E] [TRT] Could not compute dimensions for (Unnamed Layer* 6) [Convolution]_output, because the network is not valid. [03/11/2022-18:22:33] [E] [TRT] Network validation failed. Build engine successfully!

freedenS · 2022-03-12T07:53:02Z

what's the size of faster.wts?
it shows the weight value has not loaded correctly

sctrueew · 2022-03-12T09:53:10Z

what's the size of faster.wts? it shows the weight value has not loaded correctly

faster.wts size is ~390 mb

I'm using mask_rcnn_R_50_FPN_3x. Does it support FPN? or I should use C4?

freedenS · 2022-03-12T10:36:11Z

what's the size of faster.wts? it shows the weight value has not loaded correctly

faster.wts size is ~390 mb

I'm using mask_rcnn_R_50_FPN_3x. Does it support FPN? or I should use C4?

it doesn't support FPN, you can only use C4

rgkannan676 · 2022-03-15T07:46:08Z

Hi @freedenS , One more doubt, can you please tell how to change the image height and width . I have an image of shape 600x875. And I change the IMAGE_HEIGHT, IMAGE_WIDTH accordingly. But still the output is of dimension INPUT_H = 480x INPUT_W = 640;

Can you please suggest what need to be changed? I tried changing INPUT_H and INPUT_W but it doesn't give any bbox detections. Can you please advise. Thank you.

rgkannan676 mentioned this issue Mar 11, 2022

tensorrtx/rcnn 在win10下make不成功 #795

Closed

sctrueew closed this as completed Mar 11, 2022

rcnn on windows #913

rcnn on windows #913

Comments

sctrueew commented Feb 16, 2022

rgkannan676 commented Mar 10, 2022

rgkannan676 commented Mar 10, 2022

rgkannan676 commented Mar 10, 2022 • edited Loading

rgkannan676 commented Mar 10, 2022

rgkannan676 commented Mar 10, 2022

sctrueew commented Mar 10, 2022

rgkannan676 commented Mar 10, 2022

freedenS commented Mar 11, 2022 • edited Loading

freedenS commented Mar 11, 2022

rgkannan676 commented Mar 11, 2022

sctrueew commented Mar 11, 2022

rgkannan676 commented Mar 11, 2022

rgkannan676 commented Mar 11, 2022

rgkannan676 commented Mar 11, 2022

rgkannan676 commented Mar 11, 2022

sctrueew commented Mar 11, 2022 • edited Loading

sctrueew commented Mar 11, 2022

rgkannan676 commented Mar 11, 2022

sctrueew commented Mar 11, 2022

rgkannan676 commented Mar 11, 2022 • edited Loading

sctrueew commented Mar 11, 2022

sctrueew commented Mar 11, 2022

freedenS commented Mar 11, 2022

sctrueew commented Mar 11, 2022 • edited Loading

sctrueew commented Mar 11, 2022 • edited Loading

rgkannan676 commented Mar 11, 2022

sctrueew commented Mar 11, 2022

freedenS commented Mar 11, 2022

sctrueew commented Mar 11, 2022 • edited Loading

rgkannan676 commented Mar 11, 2022

sctrueew commented Mar 11, 2022

freedenS commented Mar 11, 2022

sctrueew commented Mar 11, 2022

sctrueew commented Mar 11, 2022

freedenS commented Mar 12, 2022 • edited Loading

sctrueew commented Mar 12, 2022

freedenS commented Mar 12, 2022

rgkannan676 commented Mar 15, 2022 • edited Loading

rgkannan676 commented Mar 10, 2022 •

edited

Loading

freedenS commented Mar 11, 2022 •

edited

Loading

sctrueew commented Mar 11, 2022 •

edited

Loading

rgkannan676 commented Mar 11, 2022 •

edited

Loading

sctrueew commented Mar 11, 2022 •

edited

Loading

sctrueew commented Mar 11, 2022 •

edited

Loading

sctrueew commented Mar 11, 2022 •

edited

Loading

freedenS commented Mar 12, 2022 •

edited

Loading

rgkannan676 commented Mar 15, 2022 •

edited

Loading