search

QVPR / Patch-NetVLAD

Code for the CVPR2021 paper "Patch-NetVLAD: Multi-Scale Fusion of Locally-Global Descriptors for Place Recognition"
MIT License
510 stars 72 forks source link

Couldn't reproduce results in the paper #67

Closed silaopi closed 1 year ago

silaopi commented 1 year ago

Hi, I am trying to reproduce the pitts30k results int the paper. I use storage.ini (which is Single-Spatial-Patch-NetVLAD (dim=128)) and made some changes:

  1. change resume path from mapillary_WPCA to pittsburgh_WPCA (pretrained model)
  2. change batchsize and cachebatchsize to 2, due to limited gpu memory

And I got following results, which I think is wrong. I am hoping you can help me solve this problem. Thank you.

====> Recall NetVLAD@1: 0.0119
====> Recall NetVLAD@5: 0.0468
====> Recall NetVLAD@10: 0.0794
====> Recall NetVLAD@20: 0.1344
====> Recall NetVLAD@50: 0.2453
====> Recall NetVLAD@100: 0.3784
====> Recall PatchNetVLAD@1: 0.0131
====> Recall PatchNetVLAD@5: 0.0562
====> Recall PatchNetVLAD@10: 0.1023
====> Recall PatchNetVLAD@20: 0.1617
====> Recall PatchNetVLAD@50: 0.2767
====> Recall PatchNetVLAD@100: 0.3784