train YOLO data set on Windows, and run on Jetson Nano (Ubuntu)

[VladimirSirbu]

Is it possible to train YOLO data set on PC (for example Windows) and then run file.weights on Nvidia Jetson Nano ?

[ricchagas]

Yes, it,s my use case Get Outlook for Android<https://aka.ms/AAb9ysg>

…

________________________________ From: VladimirSirbu ***@***.***> Sent: Tuesday, April 5, 2022 5:08:30 PM To: AlexeyAB/darknet ***@***.***> Cc: Subscribed ***@***.***> Subject: [AlexeyAB/darknet] train YOLO data set on Windows, and run on Jetson Nano (Ubuntu) (Discussion #8451) Is it possible to train YOLO data set on PC (for example Windows) and then run file.weights on Nvidia Jetson Nano ? — Reply to this email directly, view it on GitHub<https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FAlexeyAB%2Fdarknet%2Fdiscussions%2F8451&data=04%7C01%7C%7Ce4352d1fc8b641094fb808da17400e63%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637847861122959539%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=uXuWnIM5cdBXaFQEOBVLzvPgpRYFLlciFSXA6G9kSuw%3D&reserved=0>, or unsubscribe<https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAJHQNCAALRH4WBC2DXHDFP3VDSMT5ANCNFSM5ST3GY6Q&data=04%7C01%7C%7Ce4352d1fc8b641094fb808da17400e63%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637847861122959539%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=ZKRy1q60aE8w4S70KsxCphAEHq1K9ia0sYAvr%2BaUfSI%3D&reserved=0>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[VladimirSirbu]

Great ! And what FPS it had when you run detection ?
For example, I want to detect real-time by camera, and I need good FPS even.

[stephanecharette]

Jetson Nano won't give you much. It is more-or-less a toy.
You can see the results here: https://www.ccoderun.ca/programming/2021-10-16_darknet_fps/

Read to the very bottom of that page to understand the numbers. The realistic numbers for Jetson Nano are < 5 FPS when processing 1920x1080 video using a 416x416 yolov4-tiny network.

[VladimirSirbu]

So, on my computer I have just AMD Radeon Graphics (GPU). I think it isn't enough to run CUDA for example.

[ricchagas]

Isn't impossible train in CPU, but will take a lot of time . Other alternative is use Google colab with GPU, Get Outlook for Android<https://aka.ms/AAb9ysg>

…

________________________________ From: VladimirSirbu ***@***.***> Sent: Tuesday, April 5, 2022 7:11:29 PM To: AlexeyAB/darknet ***@***.***> Cc: Ricardo Lucas Chagas ***@***.***>; Comment ***@***.***> Subject: Re: [AlexeyAB/darknet] train YOLO data set on Windows, and run on Jetson Nano (Ubuntu) (Discussion #8451) So, on my computer I have just AMD Radeon Graphics (GPU). I think it isn't enough to run CUDA for example. — Reply to this email directly, view it on GitHub<https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FAlexeyAB%2Fdarknet%2Fdiscussions%2F8451%23discussioncomment-2511594&data=04%7C01%7C%7C45551b88fe9a4695091708da17513c1f%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637847934908139556%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=%2BK7hpa4qh5XTnZ8TXpYiWLdyHHHfOSUzJFCjohpLzJM%3D&reserved=0>, or unsubscribe<https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAJHQNCEUUJR2HHGPTZDLWULVDS3BDANCNFSM5ST3GY6Q&data=04%7C01%7C%7C45551b88fe9a4695091708da17513c1f%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637847934908139556%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=ZLiQDTTs1BjTpLJcl5tvBIr3SfT3nfQv9F5F%2Bmhk8uc%3D&reserved=0>. You are receiving this because you commented.Message ID: ***@***.***>

[ricchagas]

In my use case I'm using a raspberry pi with a neural stick and I got about 7 fps . But depending of your input video resolution. For Jetson there are a Nvidia sdk called deep stream , it's support yolo out of box , Get Outlook for Android<https://aka.ms/AAb9ysg>

…

________________________________ From: VladimirSirbu ***@***.***> Sent: Tuesday, April 5, 2022 6:50:33 PM To: AlexeyAB/darknet ***@***.***> Cc: Ricardo Lucas Chagas ***@***.***>; Comment ***@***.***> Subject: Re: [AlexeyAB/darknet] train YOLO data set on Windows, and run on Jetson Nano (Ubuntu) (Discussion #8451) Great ! And what FPS it had when you run detection ? For example, I want to detect real-time by camera, and I need good FPS even. — Reply to this email directly, view it on GitHub<https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FAlexeyAB%2Fdarknet%2Fdiscussions%2F8451%23discussioncomment-2511522&data=04%7C01%7C%7Cf5c9fe6ee64a47d245d108da174e5013%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637847922361711780%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=mkD%2BrcHAtP6G6KCD4B9vYwmEeyc7lC26HeAj2jbIcx4%3D&reserved=0>, or unsubscribe<https://nam12.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAJHQNCCRWMVOQTERFYHQPCLVDSYSTANCNFSM5ST3GY6Q&data=04%7C01%7C%7Cf5c9fe6ee64a47d245d108da174e5013%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637847922361711780%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000&sdata=BB9Kxo11gDT2tSpACaxTfYHrWyKFy0mcoRWxO4FAywI%3D&reserved=0>. You are receiving this because you commented.Message ID: ***@***.***>