I am training a custom object detection model using Yolov3 and darknet. I initially started with google colab and now I exhausted the capacity on colab. Hence I migrated to AWS to access the GPU. While working in Colab, I had no major issues. The make command worked perfectly, but now when I try to resume the training of the model on the AWS server, I am getting an error building the darknet
I have modified the Makefile as:
GPU = 1
CUDNN = 1
OpenCV = 1
When I set GPU=0,it works fine, but when I set GPU=1, I get the following error. Please find the attached screenshot.
After reading other blogs, I even downgraded the CUDA version from 10.0 to 8.0.
Please help me solve this issue. Thanks in advance.
Please click on this below link to find the screenshot of the error.
AWS-Makefile-Error
The problem was mismatch of CUDA version with darknet. After upgrading to the latest version - CUDA 11.3, my problem was solved!
Related
I am very new to the vulkan API and have been having problems with getting it working. When I run the vulkan SDK examples or attempt to run my own code, vulkan is unable to find any supported devices. I have attempted to completely reinstall my display drivers, a complete reinstall of the vulkan SDK, a manual install of the vulkan runtime and tried installing the beta version of the nvidia vulkan drivers. I am running an Nvidia GeForce GTX 1660 Ti which should support vulkan just fine. The driver version is 451.79 and the vulkan SDK version is 1.2.148.0.
When it comes to using my own code, I am able to succesfully create an instance and setup debug callbacks. The error comes in as soon as I attempt to use vkEnumeratePhysicalDevices. The callback prints out this error:
Vulkan Callback: [Loader Message] code: 0 : setupLoaderTrampPhysDevs: Failed during dispatch call of 'vkEnumeratePhysicalDevices' to lower layers or loader to get count.
Along with that error I immediately check if the device count is 0, which ends up being true.
I feel like I have a problem with my vulkan RT or vulkan SDK install but I can't figure out what it is. Any help is greatly appreciated!
EDIT: I have also tried running vulkaninfo and that does successfully find my GPU and lists information about it.
I`m facing issue that with OSDK-ROS 3.7 and 3.8, I could not run demo_flight_control sample correctly
After api auto taking off in the simulator, the drone just stuck there. And there is command being published to the flight_control ENU topic.
The GPS based demo_mission has no issue. It can auto fly to the coordinate.
The hardware and firmware version I`m using is
STATUS/1 # parseDroneVersionInfo, L729: Hardware = A3
STATUS/1 # parseDroneVersionInfo, L730: Firmware = 3.3.8.47
The minimal sample for reproducing the issue is from here
https://github.com/dji-sdk/Onboard-SDK-ROS/tree/3.8/dji_sdk_demo/src
The flight_control issue is the same for OSDK (option b) and OSDK-ROS (default)
For Previous 3.3 3.5 3.6 I never encounter this before. I`ve send report to dev#dji.com. No response yet
I solve the problem by myself
On https://github.com/dji-sdk/Onboard-SDK-ROS/tree/3.7 and 3.8 reqires at least 1.7.6
But I can't find it 1.7.6 in the DJI assistant 2 firmware only 1.7.7 and 1.7.5. So I decided to use 1.7.7
But 1.7.7 does only work for GPS waypoint mission and does not support setpoint_generic based control
So the solution is to use 1.7.5 to get both GPS based mission and ros topic based velocity control working.
Below this version, the version control in the DJI system is hell with randomly labeled version. So you have to keep trial and error to get things up
Below is DJI Email response. Pretty useless but will provide some cue for others to debug the similar issue
Bruce Cheung （DJI）
Dear Shenghai,
Thank you for contacting DJI.
We cannot reproduce the problem you mentioned by using A3 with OSDK3.7 and 3.8
Can you try to take control authority directly? using /dji_sdk/sdk_control_authority
Also try publishing the /dji_sdk/flight_control_setpoint_ENUposition_yaw control command, do not use the sample.
Thank you for your understanding and support, hope you a delighted life.
Best Regards,
DJI SDK Support Team
Hi Bruce
I tried what you mentioned., use
FIrst launch sdk.launch
Then, rosservical call /dji_sdk/sdk_control_authority 1
And write a simple empty C++ script to publish to /dji_sdk/flight_control_setpoint_generic.
The result is the same. the drone is not moving in the DJI assistant2 simulator.
My theory is the A3 firmware version might be too high on GitHub https://github.com/dji-sdk/Onboard-SDK-ROS/tree/3.7
it says 3.7 and 3.8 support
A3/A3 Pro 1.7.6.0
But I can't find it 1.7.6 in the DJI assistant 2 firmware only 1.7.7 and 1.7.5. So I decided to use 1.7.7
Will this be the issue?
Its 11 pm in Singapore, I ll take two videos of sample and custom method tmr morning. I`ll upload the video to youtube. you can see from there.
Regards
Shenghai Yuan
I have a retrained_graph.pb file which I want to convert to tensorflow lite.I tried to do it with toco in Docker but it seems that there is no module such called toco in my system.I have the latest version of tensorflow installed.Can someone please tell me how to do it?
I am running on Windows 10 64bit
I'm the community manager of Heartbeat (an online pub that looks at the intersection of mobile and ML. One of our contributors just published a tutorial that addresses this exact issue. Hope it's helpful!
https://heartbeat.fritz.ai/working-through-a-tensorflow-lite-tutorial-on-windows-10-e27ee0e8b8cc
I'm trying to install Unity 3D V 5.3.1 on my Mac. I was able to download the UnityDownloadAssistant.dmg file, however when I run the installer, it fails Download Assistant Failure Error Screenshot. Please see the screenshot.
I've been stuck on this for the last 10 days, I've tried with different network connections as well. I went ahead and checked their old downloads as suggested by some other people on the internet. But all their old links are broken.
Anyone else come across this issue and any solutions please??
I've found this helpful tutorial on web
http://blog.rayapps.com/2009/09/14/how-to-install-oracle-database-10g-on-mac-os-x-snow-leopard/
I've followed all steps, but, I've a problem with netca run. When I'll start it, crash with this error:
Invalid memory access of location 00000014 eip=11069523
/opt/oracle/product/10.2.0/db_1/jdk/jre/bin/java: line 2: 10323 Bus error /System/Library/Frameworks/JavaVM.framework/Versions/1.4.2/Home/bin/java -d32 -Xbootclasspath/a:/opt/oracle/product/10.2.0/db_1/jdk/jre/lib/ext:/opt/oracle/product/10.2.0/db_1/jdk/lib/ext $*
and the error message on popup show that the problem is with libnjni10.jnilib plugin. On snow leopard there is only Java 1.6 but I've installed also JDK1.5 and I've changed the symbolic link for 1.4.2 to run the 1.5
Any idea?
thanks a lot!
Andrea
Alex Gorbachev's install guide suggests a couple of different ways of resolving this problem - editing runinstaller or installing JDK 1.4.2. Check it out.
Try Using VMWare
Is this for development purposes? I've been happy with installing Oracle on a VMWare Linux instance. One nice feature is that I can prep my database as I like, take a VMWare snapshot, and roll back to the saved snapshot as necessary.