search

yuminsuh / part_bilinear_reid

Code for ECCV2018 paper: Part-Aligned Bilinear Representations for Person Re-Identification
141 stars 30 forks source link

issue in handling complex float numbers #18

Open abhinine4 opened 2 years ago

abhinine4 commented 2 years ago

We are facing issue in calculating pdist() as it is getting complexFloat inputs and this is not handled in later versions of pytorch. How should we handle this problem,

bhosalems commented 2 years ago

This issue arises as code base uses torch.fft() method which is deprecated from pytorch 1.7. In particular, this method is used in compact bilinear pooling, which uses *fft() methods. I am able to bypass this issue with the implementation of compact Bilinearpooling here - https://github.com/DeepInsight-PCALab/CompactBilinearPooling-Pytorch