can't train with cpu only

[SnowyJune973]

Because of some certain reason, I want to train the model with small data on a PC with no GPU. But Python gave me an AssertionError as I found here.

Traceback (most recent call last):
  File "/home/snowyjune/local/faster-rcnn.pytorch/faster-rcnn.pytorch/trainval_net.py", line 336, in <module>
    loss.backward()

  File "/home/snowyjune/anaconda3/lib/python3.6/site-packages/torch/tensor.py", line 93, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph)

  File "/home/snowyjune/anaconda3/lib/python3.6/site-packages/torch/autograd/__init__.py", line 89, in backward
    allow_unreachable=True)  # allow_unreachable flag

  File "/home/snowyjune/local/faster-rcnn.pytorch/faster-rcnn.pytorch/lib/model/roi_align/functions/roi_align.py", line 38, in backward
    assert((self.feature_size is not None) and (grad_output.is_cuda))

AssertionError

It seems that the program is designed not to support training with CPU. Is that true?

[SnowyJune973]

If not, how could I run the code properly with CPU only? Thanks so much!

[jwyang]

@SnowyJune973 please use roi_pooling instead of roi_align

[SnowyJune973]

@SnowyJune973 please use roi_pooling instead of roi_align

I added cfg.POOLING_MODE = 'pool' under Line 260 of trainval_net.py, and ran trainval_net.py again with arguments '--bs 1', but it still didn't work.

Python still gave me the same AssertionError, but this time it came from roi_pool.py

Traceback (most recent call last):
  File "trainval_net.py", line 333, in <module>
    loss.backward()
  File "/home/cch/anaconda3/lib/python3.6/site-packages/torch/tensor.py", line 93, in backward
    torch.autograd.backward(self, gradient, retain_graph, create_graph)
  File "/home/cch/anaconda3/lib/python3.6/site-packages/torch/autograd/__init__.py", line 89, in backward
    allow_unreachable=True)  # allow_unreachable flag
  File "/home/cch/faster-rcnn.pytorch/lib/model/roi_pooling/functions/roi_pool.py", line 31, in backward
    assert(ctx.feature_size is not None and grad_output.is_cuda)
AssertionError

[jwyang]

@SnowyJune973 I will change some code, will let you know when it works.

[SnowyJune973]

@SnowyJune973 I will change some code, will let you know when it works.

I'm so sorry for the inconvenience.

[jihwan1008]

@SnowyJune973 Sorry to bother, but did you succeeded in 'sh make.sh'? Didn't you get any cffi errors?

[kris-singh]

I have the same issue here. I just want to check the code for working can i use pretrained model for checking i don't care about the results.
Looked at the issue more carefully you can't really train using any of the roi pooling as they only implement the backward pass for cuda. But i feel testing would still work.

[kangkang59812]

@SnowyJune973 Does it work when traing a small dataset with CPU only?

[simomagi]

I have the same problem. For some testing purpose, i'm not able to train faster R-CNN on a CPU only machine. I have cffi errors when i run "sh make.sh". How can i solve it?

[ControllableGeneration]

Same as @jihwan1008 and @simomagi, I was working in MacOS with CPU only and trying to run sh make.sh but got a cffi error (cffi.VerificationError: LinkError: command 'gcc' failed with exit status 1). What can I do with it? Issue #678

Same as @jihwan1008 and @simomagi, I was working in MacOS with CPU only and trying to run sh make.sh but got a cffi error (cffi.VerificationError: LinkError: command 'gcc' failed with exit status 1).
please advice

[EMCP]

@mekong0404 your mac likely doesn't have gcc installed

see https://stackoverflow.com/questions/20410587/how-to-find-gcc-version-on-mac

it's basically clang , not gcc .. and quite old

sorry for trouble , but when I check for gcc by putting gcc --version

onfigured with: --prefix=/Library/Developer/CommandLineTools/usr --with-gxx-include-dir=/Library/Developer/CommandLineTools/SDKs/MacOSX10.14.sdk/usr/include/c++/4.2.1
Apple LLVM version 10.0.1 (clang-1001.0.46.4)
Target: x86_64-apple-darwin18.7.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin

what should I do

[ControllableGeneration]

It says clang there. I think that's what he meant. You probably cannot get CPU-only mode running. I am now turning into clusters