search

thomasverelst / dynconv

Code for Dynamic Convolutions: Exploiting Spatial Sparsity for Faster Inference (CVPR2020)
https://arxiv.org/abs/1912.03203
126 stars 14 forks source link

/annot/valid.json is missing #8

Open zhang-pengyu opened 3 years ago

zhang-pengyu commented 3 years ago

Thanks for your inspiring work on dynamic convolution. I am interested in this exciting work and try to test the efficiency of dynconv. I run the test code for pose estimation and met a exception: No such file or directory: '$mpii_root/annot/valid.json'. I wonder how can I get this file.

BTW, when I set the stride = 2 for conv3x3_dw. I often crashed with 'THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1565272279342/work/aten/src/THC/THCCachingHostAllocator.cpp line=296 error=77 : an illegal memory access was encountered'. Could you help me debug those issues? Thanks in advance.

Pengyu Zhang.

thomasverelst commented 3 years ago

The valid.json is part of the MPII dataset. Did you download that one and is $mpii_root pointing to the right location?

For the CUDA error: did you set stride=2 for the MaskUnit too?

zhang-pengyu commented 3 years ago

I downloaded the dataset via http://human-pose.mpi-inf.mpg.de/#download, and the extension of annotation file is '.mat'. Could you point out the relative path of 'valid.json'

Since I did not test on pose estimation task, I just borrow the raw code and I check the stride is the same as conv layer. I will appreciate if you can provide some ideas on how to debug it. Thanks in advance.

zhang-pengyu commented 3 years ago

I found that the error stems from the cupy code 'masked_conv' and it does not always appear. It seems strange. I have no idea for debugging it. Have you met this issue?