search

dragonlee258079 / DMT

Code release for the CVPR 2023 paper "Discriminative Co-Saliency and BG Mining Transformer for Co-Salient Object Detection" by Long Li, Junwei Han, Ni Zhang, Nian Liu*, Salman Khan, Hisham Cholakkal, Rao Muhammad Anwer, Fahad Shahbaz Khan
MIT License
18 stars 2 forks source link

复现不出论文精度 #5

Open zhending111 opened 7 months ago

zhending111 commented 7 months ago

您好，请问是还有其他什么训练策略么？训练了几次均复现不出论文中的精度，S measure 连69都没超过

dragonlee258079 commented 7 months ago

你好，我的训练策略都在train.py代码里了，至于复现不出精度，你可以检查一下环境是否一致，下载的训练数据组合是否正确

zhending111 commented 7 months ago

您好，那个syn总共有几个呢?

zhending111 commented 7 months ago

我就好像有两个syn，一个是add_navie另一个是add_navie_reverse2

dragonlee258079 commented 7 months ago

synthesis strategy 是 follow CADC（Summarize and search: Learning consensus-aware dynamic convolution for co-saliency detection） 方法的，合成策略就是正向合成和反向合成，一共是两个

zhending111 commented 7 months ago

synthesis strategy 是 follow CADC（Summarize and search: Learning consensus-aware dynamic convolution for co-saliency detection） 方法的，合成策略就是正向合成和反向合成，一共是两个

好的，我去检查一下环境吧，训练数据应该是没问题，感谢及时回复

ZhengPeng7 commented 3 weeks ago

Hi, @dragonlee258079, so, what is the performance of DMT without the data synthesis strategy in CADC? Since most CoSOD methods didn't use that, the performance without it can be instrumental for comparison. Thanks!

dragonlee258079 commented 3 weeks ago

@ZhengPeng7 Thank you for your suggestion. I will provide the performance without the data synthesis strategy soon.