Load extention fault on jetson tx2

[omeralierdemir]

hi everyone, I installed this repo on jetson tx2. After I run project, I annotated target area . then system was freezed and I did not get any mistake compiler error. I have waited for along time but I did not take any response. I deep dive and debug. The code which is following the bottom not runing:

_prroi_pooling = load_extension( '_prroi_pooling', [pjoin(root_dir, 'prroi_pooling_gpu.c'), pjoin(root_dir, 'prroi_pooling_gpu_impl.cu')], verbose=True) (functional.py was included this code)

[goutamgmb]

Hi,

Unfortunately we haven't tested the code on jetson tx2. Thus I'm not sure whether the PrPool module can be compiled on jetson tx2 or not. You could try running a variant of ATOM which does not use the PrPool module using the parameter setting multiscale_no_iounet.

[franva]

hi @goutamgmb and @omeralierdemir

Do we have any update for this topic?

With more and more edge devices coming with AI built-in, running AI models on edge devices has become more popular.

I have a project which also requires to run SOT on RPi 4 with 4B. I have the OAK-D camera which integrates the Intel Movidius MyriadX VPU, I also have a Intel Myriad Stick 2 which I would like to utilize in my project.

I would really appreciate if someone could provide some update about running the fantastic PyTracking(I have tried the ATOM and DiMP-18 myself and they are amazing~!) on any edge devices.

Many thanks~!

[goutamgmb]

Hi,

As mentioned in issue #281, I would suggest to first try if you can install all the dependencies of PyTracking except the Precise ROI pooling package.

[Ahsanr312]

Hi,

Unfortunately we haven't tested the code on jetson tx2. Thus I'm not sure whether the PrPool module can be compiled on jetson tx2 or not. You could try running a variant of ATOM which does not use the PrPool module using the parameter setting multiscale_no_iounet.

I have tried it with Pytorch 1.8 and successfully run the following Trackers (Dimp, PrDimp, KeepTrack) on Jetson Tx2 and I believe with Pytorch 1.7 I would be able to use ATOM too. I am currently facing problem using KYS while installing its Spatial Correlation Sampler, its unable to install with Pytorch 1.8 yet. Secondly, LWL is not working on run_video.py, looks like the support for LWL is still not given in run_video.py

[franva]

thanks @Ahsanr312 for the update.

Do you know whether the success on jetson could be replicate on Raspberry Pi 4 ?

[Ahsanr312]

It's all about trying it. I haven't done it so far so can't say much, but I believe you will be able to do it.