search

princeton-vl / pytorch_stacked_hourglass

Pytorch implementation of the ECCV 2016 paper "Stacked Hourglass Networks for Human Pose Estimation"
BSD 3-Clause "New" or "Revised" License
465 stars 94 forks source link

Error when running on GPU #8

Closed david-wb closed 4 years ago

david-wb commented 4 years ago

Hi,

I'm trying to run a slightly modified version of this model on the gpu, but seeing an error that was also reported in #6 

in _max_pool2d
    input, kernel_size, stride, padding, dilation, ceil_mode)
RuntimeError: max_pool2d_with_indices_out_cuda_frame failed with error code 0

Process finished with exit code 1

The problem occurs in the max pooling layer. The user in #6 worked around it by downgrading to pytorch 1.0, but that seems unsatisfactory.

Have you tested this model on the GPU and observed this error?

Thanks!

david-wb commented 4 years ago

I noticed this happens when the number of channels (inp_dim) is too large. I reduced it from 256 to 128 and the problem went away.

david-wb commented 4 years ago

I will chalk this up to an unsupported version of pytorch. The workaround is to reduce number of features (inp_dim).