RuntimeError: Expected object of backend CUDA but got backend CPU for argument #2 'other'

[linchunmian]

Hi, I followed the installation and get_start files to complete necessary running environment, and this bug happened. How should I solve this, and please help me !
Thanks!

[tianweiy]

Seems that some value need to be converted to cuda. You can just add a cuda() at that line ?

[linchunmian]

Thanks. It confuses me that I just clone the project without any modifications, and the environment is ok. Also, I could run other 3D detection model, e.g. PV-RCNN with corresponding config file, and the result is normal. You know, the most difference bwtween centerpoint_rcnn and pv-rcnn is the part of DenseHead. I have checked the variances that maybe occur error in centerpoint_head.py carefully, but unfortunately the bug still exists.Could you please give me a hand? And any instructions would be appreciated.Regards. ---- Replied Message ----FromTianwei ***@***.***>Date05/25/2021 21:34 ***@***.***> ***@***.******@***.***>SubjectRe: [tianweiy/CenterPoint-KITTI] RuntimeError: Expected object of backend CUDA but got backend CPU for argument #2 'other' (#3) Seems that some value need to be converted to cuda. You can just add a cuda() at that line ? —You are receiving this because you authored the thread.Reply to this email directly, view it on GitHub, or unsubscribe. [ { ***@***.***": "http://schema.org", ***@***.***": "EmailMessage", "potentialAction": { ***@***.***": "ViewAction", "target": "#3 (comment)", "url": "#3 (comment)", "name": "View Issue" }, "description": "View this Issue on GitHub", "publisher": { ***@***.***": "Organization", "name": "GitHub", "url": "https://github.com" } } ]

[tianweiy]

I mean adding a cuda() to this line you referenced #2 (comment)

Does this work? If not please paste the full log, this is easier than sentences for debugging

[linchunmian]

I mean adding a cuda() to this line you referenced #2 (comment)

Does this work? If not please paste the full log, this is easier than sentences for debugging

Thanks. The error information is as follows:

Where should cuda() be added?

[tianweiy]

either center_int or feature_map_size

[tianweiy]

like

center_int[0].cuda() < feature_map_size[0].cuda() ...

[linchunmian]

like

center_int[0].cuda() < feature_map_size[0].cuda() ...

Many thanks. As you said, I have added cuda() behind the variance of 'feature_map_size', and new type problem happens:

How to solve it? By the way, adding cuda() after center_int, e.g center_int().cuda() does not make effect.

[tianweiy]

try this

if not (0 <= center_int[0] < feature_map_size[0].cuda().int()
      and 0 <= center_int[1] < feature_map_size[1].cuda().int()):

[linchunmian]

I convert the type of feature_map_size to int32 by adding .to(torch.int32).
It seems this bug has passed, and the code run normally.
Many thanks for your helps.

[bi-qiuqiu]

I convert the type of feature_map_size to int32 by adding .to(torch.int32). It seems this bug has passed, and the code run normally. Many thanks for your helps.
Did you change all feature_map_size in this file to feature_map_size.to(torch.int32)?I changed line 269 to if not (0 <= center_int[0] < feature_map_size[0].cuda().int()，but this problem arises.
and 0 <= center_int[1] < feature_map_size[1].cuda().int()):