CVUSA training exception

wtyhub / LPN

Pytorch implementation of Each Part Matters: Local Patterns Facilitate Cross-view Geo-localization https://arxiv.org/abs/2008.11646

MIT License

75 stars 13 forks source link

CVUSA training exception #2

Closed Allen-lz closed 3 years ago

Allen-lz commented 3 years ago

hi,Thanks for your work！ But when I was training CVUSA, the loss was always very large, and the accuracy rate was very low. I have processed the dataset according to prepare_cvusa.py

-- Epoch 33/189

-- train Loss: 122.1055 Satellite_Acc: 0.0339 Street_Acc: 0.0411

-- Training complete in 450m 17s

wtyhub commented 3 years ago

Don‘t worry. When the epoch is 80 and the learing rate becomes 0.1*lr, the accuracy will improve.

Allen-lz commented 3 years ago

Thank you for your prompt reply, it is indeed the case.