Can't use Jetson-inference on Drive PX2

anon18001062 · July 11, 2018, 7:43am

I’m trying to use Jetson-inference on Drive PX2(Auto Chauffeur).
Following the site below, I’ve done to “make” Jetson-inference on DRIVE PX2.
https://github.com/dusty-nv/jetson-inference

However, trying to do an “orange classify” results in the following log and cannot be classified

I found a similar issue on the site below, but I could not solve it.
<https://devtalk.nvidia.com/default/topic/1023362/general/issues-with-tensorrt-on-drive-px2 />

Is it possible to use Jetson inference on DrivePX platform ?

$ ./imagenet-console orange_0.jpg output_0.jpg

imagenet-console args
 (3): 0 [./imagenet-console] 1 [orange_0.jpg] 2 [output_ 0.jpg] 

imageNet - loading classification network model from:
         - prototxt networks / googlenet.prototxt
         - model networks / bvlc_googlenet.caffemodel
         - class_labels networks / ilsvrc12_synset_words.txt
         - input_blob 'data'
         - output_blob 'prob'
         - batch_size 2

[GIE] TensorRT version 3.0.2, build 3002 [GIE] attempting to open cache file networks / bvlc_googlenet.caffemodel.2.tensorcache
[GIE] loading network profile from cache ...
networks / bvlc_googlenet.caffemodel.2.tensorcache
nvrm_gpu: Bug 200215060 workaround enabled.
[GIE] platform does not have FP16 support.
[GIE] networks / bvlc_googlenet.caffemodel loaded [GIE] CUDA engine context initialized with 2 bindings [GIE] networks / bvlc_googlenet.caffemodel input binding index: 0 [GIE] networks / bvlc_googlenet.caffemodel input dims (b = 2 c = 3 h = 224 w = 224) size = 1204224 [cuda] cudaAllocMapped 1204224 bytes, CPU 0x1053e0000 GPU 0x1053e0000 [GIE] networks / bvlc_googlenet.caffemodel output 0 prob binding index: 1 [GIE] networks / bvlc_googlenet.caffemodel output 0 prob mds (b = 2 c = 1000 h = 1 w = 1) size = 8000 [cuda] cudaAllocMapped 8000 bytes, CPU 0x1055e0000 GPU 0x1055e0000 networks / bvlc_googlenet.caffemodel initialized.
[GIE] networks / bvlc_googlenet.caffemodel loaded imageNet - loaded 1000 class info entries networks / bvlc_googlenet.caffemodel initialized.
loaded image orange_0.jpg (1920 x 1920) 58982400 bytes [cuda] cuda AllocMapped 58982400 bytes, CPU 0x1057e0000 GPU 0x1057e0000 [cuda] cudaGetLastError () [cuda] no kernel image is available for execution on the device (error
48) (hex 0 x 30)
[cuda] /home/nvidia/jetson-inference/imageNet.cu:115
[cuda] cudaPreImageNetMean ((float 4 *) rgba, width, height, mInputCUDA, mWidth, mHeight, make_float 3 (104.0069879317889 f, 116.66876761696767 f,
122.6789143406786 f))
[cuda] no kernel image is available for execution on the device (error
48) (hex 0 x 30)
[cuda] /home/nvidia/jetson-inference/imageNet.cpp:296
imageNet :: Classify () - cudaPreImageNetMean failed
imagenet-console: failed to classify 'orange_ 0.jpg' (result = -1)

shutting down ...

AastaLLL · July 12, 2018, 5:57am

Hi,

CUDA error 48 means no available kernel image for the device.
This usually occurs when a user doesn’t generate code with the corresponding device configuration.

Jetson_inference is a sample for Jetson platform and doesn’t includes the GPU architecture of PX2.
Please modify the configuration with following change to add PX2 GPU support.

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 3026acc..82b4bc5 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -33,7 +33,7 @@ set(
        CUDA_NVCC_FLAGS
        ${CUDA_NVCC_FLAGS}; 
     -O3 
-       -gencode arch=compute_53,code=sm_53
+       -gencode arch=compute_61,code=sm_61
        -gencode arch=compute_62,code=sm_62
 )

Thanks.

madergona · July 12, 2018, 9:31am

Hi,

may I asked you, how you were able to compile the jetson-inference project for the Drive PX2?
I am trying this, too, but I am not able to get it run.

Thank you for your help!
Best regards

anon18001062 · July 12, 2018, 11:51pm

Hi, AastaLLL.

After modifying CMakeLists.txt according to your advice and cmake it again,
the console command of classify was executed.
Thank you for your advice!

anon18001062 · July 13, 2018, 4:05am

Next step.

I tried using webcam to recognize by jetson-inference command “./Imagenet-camera”.
However, the image of the web camera was black.
Do I need to set something?

I’m using Logicool C922 web camera.
I confirmed that the web camera can operate by using the command “/usr/local/driveworks/bin/sample_camera_usb --device=0”.

The log when executing the command “./imagenet-camera” is as follows

$ ./imagenet-camera
imagenet-camera
  args (1):  0 [./imagenet-camera]  

[gstreamer] initialized gstreamer, version 1.8.2.0
[gstreamer] gstreamer decoder pipeline string:
v4l2src device=/dev/video0 ! video/x-raw, width=(int)1280, height=(int)720, format=RGB ! videoconvert ! video/x-raw, format=RGB ! videoconvert !appsink name=mysink

imagenet-camera:  successfully initialized video device
    width:  1280
   height:  720
    depth:  24 (bpp)


imageNet -- loading classification network model from:
         -- prototxt     networks/googlenet.prototxt
         -- model        networks/bvlc_googlenet.caffemodel
         -- class_labels networks/ilsvrc12_synset_words.txt
         -- input_blob   'data'
         -- output_blob  'prob'
         -- batch_size   2

[GIE]  TensorRT version 3.0.2, build 3002
[GIE]  attempting to open cache file networks/bvlc_googlenet.caffemodel.2.tensorcache
[GIE]  loading network profile from cache... networks/bvlc_googlenet.caffemodel.2.tensorcache
nvrm_gpu: Bug 200215060 workaround enabled.
[GIE]  platform does not have FP16 support.
[GIE]  networks/bvlc_googlenet.caffemodel loaded
[GIE]  CUDA engine context initialized with 2 bindings
[GIE]  networks/bvlc_googlenet.caffemodel input  binding index:  0
[GIE]  networks/bvlc_googlenet.caffemodel input  dims (b=2 c=3 h=224 w=224) size=1204224
[cuda]  cudaAllocMapped 1204224 bytes, CPU 0x1053e0000 GPU 0x1053e0000
[GIE]  networks/bvlc_googlenet.caffemodel output 0 prob  binding index:  1
[GIE]  networks/bvlc_googlenet.caffemodel output 0 prob  dims (b=2 c=1000 h=1 w=1) size=8000
[cuda]  cudaAllocMapped 8000 bytes, CPU 0x1055e0000 GPU 0x1055e0000
networks/bvlc_googlenet.caffemodel initialized.
[GIE]  networks/bvlc_googlenet.caffemodel loaded
imageNet -- loaded 1000 class info entries
networks/bvlc_googlenet.caffemodel initialized.
default X screen 0:   2960 x 1050
[OpenGL]  glDisplay display window initialized
[OpenGL]   creating 1280x720 texture
loaded image  fontmapA.png  (256 x 512)  2097152 bytes
[cuda]  cudaAllocMapped 2097152 bytes, CPU 0x1057e0000 GPU 0x1057e0000
[cuda]  cudaAllocMapped 8192 bytes, CPU 0x1055e2000 GPU 0x1055e2000
[gstreamer] gstreamer transitioning pipeline to GST_STATE_PLAYING
[gstreamer] gstreamer changed state from NULL to READY ==> mysink
[gstreamer] gstreamer changed state from NULL to READY ==> videoconvert1
[gstreamer] gstreamer changed state from NULL to READY ==> capsfilter1
[gstreamer] gstreamer changed state from NULL to READY ==> videoconvert0
[gstreamer] gstreamer changed state from NULL to READY ==> capsfilter0
[gstreamer] gstreamer changed state from NULL to READY ==> v4l2src0
[gstreamer] gstreamer changed state from NULL to READY ==> pipeline0
[gstreamer] gstreamer changed state from READY to PAUSED ==> videoconvert1
[gstreamer] gstreamer changed state from READY to PAUSED ==> capsfilter1
[gstreamer] gstreamer changed state from READY to PAUSED ==> videoconvert0
[gstreamer] gstreamer changed state from READY to PAUSED ==> capsfilter0
[gstreamer] gstreamer stream status CREATE ==> src
[gstreamer] gstreamer changed state from READY to PAUSED ==> v4l2src0
[gstreamer] gstreamer changed state from READY to PAUSED ==> pipeline0
[gstreamer] gstreamer stream status ENTER ==> src
[gstreamer] gstreamer msg new-clock ==> pipeline0
[gstreamer] gstreamer msg stream-start ==> pipeline0
[gstreamer] gstreamer changed state from PAUSED to PLAYING ==> videoconvert1
[gstreamer] gstreamer changed state from PAUSED to PLAYING ==> capsfilter1
[gstreamer] gstreamer changed state from PAUSED to PLAYING ==> videoconvert0
[gstreamer] gstreamer changed state from PAUSED to PLAYING ==> capsfilter0
[gstreamer] gstreamer changed state from PAUSED to PLAYING ==> v4l2src0

imagenet-camera:  camera open for streaming
[gstreamer] gstreamer decoder onPreroll
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x1059e0000 GPU 0x1059e0000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x105c90000 GPU 0x105c90000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x105f40000 GPU 0x105f40000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x1061f0000 GPU 0x1061f0000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x1064a0000 GPU 0x1064a0000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x106750000 GPU 0x106750000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x106a00000 GPU 0x106a00000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x106cb0000 GPU 0x106cb0000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x106f60000 GPU 0x106f60000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x107210000 GPU 0x107210000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x1074c0000 GPU 0x1074c0000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x107770000 GPU 0x107770000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x107a20000 GPU 0x107a20000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x107cd0000 GPU 0x107cd0000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x107f80000 GPU 0x107f80000
[cuda]  cudaAllocMapped 2764800 bytes, CPU 0x108230000 GPU 0x108230000
[cuda]   gstreamer camera -- allocated 16 ringbuffers, 2764800 bytes each
[gstreamer] gstreamer changed state from READY to PAUSED ==> mysink
[gstreamer] gstreamer msg async-done ==> pipeline0
[gstreamer] gstreamer changed state from PAUSED to PLAYING ==> mysink
[gstreamer] gstreamer changed state from PAUSED to PLAYING ==> pipeline0
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x1084e0000 GPU 0x1084e0000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x1092f0000 GPU 0x1092f0000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x10a100000 GPU 0x10a100000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x10af10000 GPU 0x10af10000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x10bd20000 GPU 0x10bd20000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x10cb30000 GPU 0x10cb30000
[gstreamer] gstreamer msg qos ==> v4l2src0
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x10d940000 GPU 0x10d940000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x10e750000 GPU 0x10e750000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x10f560000 GPU 0x10f560000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x110370000 GPU 0x110370000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x111180000 GPU 0x111180000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x111f90000 GPU 0x111f90000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x112da0000 GPU 0x112da0000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x113bb0000 GPU 0x113bb0000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x1149c0000 GPU 0x1149c0000
[cuda]  cudaAllocMapped 14745600 bytes, CPU 0x1157d0000 GPU 0x1157d0000
[cuda]   gstreamer camera -- allocated 16 RGBA ringbuffers
class 0475 - 0.283953  (car mirror)
class 0605 - 0.014470  (iPod)
class 0632 - 0.010347  (loudspeaker, speaker, speaker unit, loudspeaker system, speaker system)
class 0662 - 0.029539  (modem)
class 0664 - 0.098564  (monitor)
class 0782 - 0.026043  (screen, CRT screen)
class 0785 - 0.363981  (seat belt, seatbelt)
class 0851 - 0.013250  (television, television system)
class 0859 - 0.040571  (toaster)
imagenet-camera:  36.39809% class #785 (seat belt, seatbelt)
class 0475 - 0.297077  (car mirror)
class 0605 - 0.010207  (iPod)
class 0632 - 0.011224  (loudspeaker, speaker, speaker unit, loudspeaker system, speaker system)
class 0662 - 0.027714  (modem)
class 0664 - 0.093532  (monitor)
class 0782 - 0.026952  (screen, CRT screen)
class 0785 - 0.378346  (seat belt, seatbelt)
class 0851 - 0.012728  (television, television system)
class 0859 - 0.037469  (toaster)

AastaLLL · July 13, 2018, 8:09am

Hi,

Please build it with this command:
https://github.com/dusty-nv/jetson-inference#cloning-the-repo

Please also remember to update the change in comment #2.
Thanks.

AastaLLL · July 13, 2018, 8:12am

Hi, h-kinumatsu

The source code to open a USB camera is designed for Jetson.
Please update the configuration here:
[url]https://github.com/dusty-nv/jetson-inference/blob/master/util/camera/gstCamera.cpp#L321[/url]

Thanks.

anon18001062 · July 17, 2018, 4:28am

Hi, AastaLLL.

Thank you for your comment.

According to comment #7,
it updates the content of the function “gstCamera::buildLaunchStr()” of gstCamera.cpp,
How do I update it?

anon18001062 · July 19, 2018, 2:10am

I will supplement the comment #5 and #8.

When executing jetson-inference command “./Imagenet-camera”,
looking at the console screen, it seems to be recognizing somewhat.
However, the image of the web camera is black.

Executing the DriveWorks command “/usr/local/driveworks/bin/sample_camera_usb --device=0”,
the web camera moves normally.
From this, I think there is no abnormality in the web camera.

I got advice at comment #7, but I do not know where to update concretely.
I tried running it with the latest gstCamera.cpp, but the result did not change.

If you have any advice, please let me know.

Best regards.

anon18001062 · July 21, 2018, 6:04am

I solved my self.
I referred to the following site.
https://github.com/dusty-nv/jetson-inference/issues/232

First change line 154 of imagenet-camera.cpp as follows.
if( !camera->ConvertRGBA(imgCUDA, &imgRGBA, true) )

Then comment out lines 197-204 of imagenet-camera.cpp and change it as follows.

// map from CUDA to openGL using GL interop
/* void* tex_map = texture->MapCUDA();

if( tex_map != NULL )
{
cudaMemcpy(tex_map, imgRGBA, texture->GetSize(), cudaMemcpyDeviceToDevice);
texture->Unmap();
} */

CUDA(cudaDeviceSynchronize());
texture->UploadCPU(imgRGBA);

After this I ran the make, I could classify it through the web camera.

I changed detectnet-camera.cpp and segnet-camera.cpp in the same way,
I could detect and segmentation through web camera.
However, segnet-camera.cpp needs to be
texture->UploadCPU(outCUDA);
instead of
texture->UploadCPU(imgRGBA);

I apologize to you for a fuss over.
Best regards.

AastaLLL · July 23, 2018, 9:17am

Thanks to share this information with us and also sorry for our late reply.