939 réponses à ce sujet

[_sebazz_]

Hi, could you please help me with the latest ISO for the RDK X5? The website is down and downloading an old image is extremely slow.

[wise_courgette_94725]

how about now. I think the web is up

[daddy_salad]

Hello, im trying to use the rdk x5for detecting uavs however, when using yolo models, like yolov8 or yolov11n its extremely slow to the point where the screen freezes. In the videos of the product its detecting humans and cars in 200 fps? Plz someone help if you can

[ed2460]

^ I've ran into the same issue. Someone recommended to use the NPU but haven't figured it out

[daddy_salad]

Yes i tried that too

[daddy_salad]

The way u use the npu or how i figured it out was by running the yolo through onxx or onxx or something

[daddy_salad]

Ultimately even with npu its kinda dog shit

[daddy_salad]

Ive been working on it non stop for 2 weeks

[aiyaz3007]

Can you explain more about how you running the inference ?

Pt file, Onnx, Bin ?
Resolution ?
Model Used ?

[daddy_salad]

Well yeah i tried pt but then vanilla pt wont use npu so then i tried onnx still kinda slow

[daddy_salad]

Resolution is like 640x something, and what do u mean by model?

[daddy_salad]

U mean like yolov8, etc?

[aiyaz3007]

Yeah

[daddy_salad]

Oh well first yolov8, but then it was giving me a hard time detecting uavs, so i tried yolov11n. And then later i tried a pre trained file called best.pt but it basically froze as soon as i ran the code. After this i thought maybe its skill issue and my code is just dog shit, but then i ran it with my laptop and it worked brilliantly

[daddy_salad]

So the code is not the issue

[daddy_salad]

Or the pt file

[daddy_salad]

It's something im missing with this device, if u can give me a hint or anything

[daddy_salad]

Based on the videos it should have no problem running these

[aiyaz3007]

Running pt or onnx won't able to access BPU of the board, convert to bin formatted model

Inference script you can get under /app dir, convert script it will be there official repo somewhere

640x480 you can able to reach 24 fps in single threat, for more optimisation reduce resolution or use multi processing

[daddy_salad]

U been able to use this to detect shit with yolo fast and smooth?