search

yxgeee / OpenIBL

[ECCV-2020 (spotlight)] Self-supervising Fine-grained Region Similarities for Large-scale Image Localization. 🌏 PyTorch open-source toolbox for image-based localization (place recognition).
https://yxgeee.github.io/projects/sfrs
MIT License
271 stars 41 forks source link

关于数据集组织的问题 #11

Closed zhilanyue closed 3 years ago

zhilanyue commented 3 years ago

你好！ 在复现你的代码的时候，执行sfrs训练（./scripts/train_sfrs_dist.sh）遇到了这样的问题： QQ截图20201217161354 问题出在(https://github.com/yxgeee/OpenIBL/blob/5ab80d65afa42ca22210c4c08983fdc156696bab/ibl/utils/data/sampler.py#L85)

在采样器中输出发现训练集的pos_list\neg_list是一样的： https://github.com/yxgeee/OpenIBL/blob/5ab80d65afa42ca22210c4c08983fdc156696bab/ibl/utils/data/sampler.py#L32 https://github.com/yxgeee/OpenIBL/blob/5ab80d65afa42ca22210c4c08983fdc156696bab/ibl/utils/data/sampler.py#L33 QQ截图20201217161459 修改过dataset.py中近邻搜索方法搜索半径，问题依然存在 https://github.com/yxgeee/OpenIBL/blob/5ab80d65afa42ca22210c4c08983fdc156696bab/ibl/utils/data/dataset.py#L28

我的数据集组织情况如下， QQ截图20201217161522 想请教一下问题所在，谢谢！

yxgeee commented 3 years ago

neg_list存的是25m以内的图，然后在sampler里会针对每个anchor的neg_list找到25m以外的negative pair，参见https://github.com/yxgeee/OpenIBL/blob/master/ibl/utils/data/sampler.py#L79

zhilanyue commented 3 years ago

neg_list存的是25m以内的图，然后在sampler里会针对每个anchor的neg_list找到25m以外的negative pair，参见https://github.com/yxgeee/OpenIBL/blob/master/ibl/utils/data/sampler.py#L79

了解，感谢！