search

haofeixu / aanet

[CVPR'20] AANet: Adaptive Aggregation Network for Efficient Stereo Matching
Apache License 2.0
521 stars 100 forks source link

Undefined Symbol Error, most likely because i am using CUDA 11 instead of the intended CUDA 10 #85

Closed shamsulfakhar closed 1 year ago

shamsulfakhar commented 1 year ago

I know I am supposed to use CUDA 10.0 for this, but my RTX 3060 requires me to use minimum CUDA 11, so here I am with a CUDA 11.8 and whenever I run bash scripts/aanet_train.sh I get this error:

File "train.py", line 8, in import nets File "/home/sf/AANET/nets/init.py", line 1, in from .aanet import AANet File "/home/sf/AANET/nets/aanet.py", line 4, in from nets.feature import (StereoNetFeature, PSMNetFeature, GANetFeature, GCNetFeature, File "/home/sf/AANET/nets/feature.py", line 5, in from nets.deform import DeformConv2d File "/home/sf/AANET/nets/deform.py", line 3, in from nets.deform_conv import DeformConv, ModulatedDeformConv File "/home/sf/AANET/nets/deform_conv/init.py", line 1, in from .deform_conv import (DeformConv, DeformConvPack, ModulatedDeformConv, File "/home/sf/AANET/nets/deform_conv/deform_conv.py", line 9, in from . import deform_conv_cuda ImportError: /home/sf/AANET/nets/deform_conv/deform_conv_cuda.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZNSt15__exception_ptr13exception_ptr10_M_releaseEv

The error message suggests that there is an undefined symbol in the file deform_conv_cuda.cpython-37m-x86_64-linux-gnu.so, which is being imported with the statement from . import deform_conv_cuda. The symbol _ZNSt15__exception_ptr13exception_ptr10_M_releaseEv is an exception handler likely a missing symbol in the library's dependencies. I assume this occur because that library was built with a different version of CUDA. Do you happen to have that library that is built using CUDA 11?

haofeixu commented 1 year ago

Unfortunately I didn't have the library with CUDA 11 since this project has finished for quite some time.

Would you be interested in taking a look at our new stereo model: https://github.com/autonomousvision/unimatch, which has better performance and is implemented in pure pytorch.