search

uber-research / UPSNet

UPSNet: A Unified Panoptic Segmentation Network
Other
649 stars 120 forks source link

Segmentation fault error #59

Open lfdeep opened 5 years ago

lfdeep commented 5 years ago

Hello!When i run train.py in coco datasets, and i only have one gpu ,it occured:

/UPSNet-master/upsnet/models/resnet_upsnet.py(125)forward() -> fcn_output = self.fcn_head(*[fpn_p2, fpn_p3, fpn_p4, fpn_p5, fcn_rois]) (Pdb) n Segmentation fault

Is the code wrong?

AugustYuan commented 5 years ago

hello, I have the same error with you, how did you deal with it? Thanks!

lfdeep commented 5 years ago

hello, I have the same error with you, how did you deal with it? Thanks!

I have solved this problem and you can: Breakpoint debugging, check if the error is in deform conv. If yes, it is because the environment is not configured, check whether the environment is compiled #6 . it is recommended that you use gcc>4.9.

AugustYuan commented 5 years ago

My gcc version is 7.2.0, and I find when the program runs to "output=train_model(*batch)" it will print error as follows, so how to deal with it?

/home/kunyuan/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py(477)call() -> result = self.forward(*input, **kwargs) (Pdb) n /home/kunyuan/anaconda3/lib/python3.6/site-packages/torch/nn/functional.py:1006: UserWarning: nn.functional.sigmoid is deprecated. Use torch.sigmoid instead. warnings.warn("nn.functional.sigmoid is deprecated. Use torch.sigmoid instead.") Segmentation fault

lfdeep commented 5 years ago

My gcc version is 7.2.0, and I find when the program runs to "output=train_model(*batch)" it will print error as follows, so how to deal with it?

/home/kunyuan/anaconda3/lib/python3.6/site-packages/torch/nn/modules/module.py(477)call() -> result = self.forward(*input, **kwargs) (Pdb) n /home/kunyuan/anaconda3/lib/python3.6/site-packages/torch/nn/functional.py:1006: UserWarning: nn.functional.sigmoid is deprecated. Use torch.sigmoid instead. warnings.warn("nn.functional.sigmoid is deprecated. Use torch.sigmoid instead.") Segmentation fault

cd upsnet/operators/_ext/deform_conv and run python import torch and import deform_conv_cuda manually, it should be no problem if your environment setup is correct.

AugustYuan commented 5 years ago

I imported deform_conv_cuda successfully, but there still "segmentation fault" error.

Flames60 commented 5 years ago

I imported deform_conv_cuda successfully, but there still "segmentation fault" error.

Same error. Have you solved it?

Flames60 commented 5 years ago

I imported deform_conv_cuda successfully, but there still "segmentation fault" error.

I solved it by upgrading gcc from 4.8.5 to 4.9.4.

AMzhanghan commented 4 years ago

Same error. Have you solved it?

AMzhanghan commented 4 years ago

I imported deform_conv_cuda successfully, but there still "segmentation fault" error.

Same error. Have you solved it?