ChunmingHe / WS-SAM

40 stars 1 forks source link

can't find scribble annotations #3

Open 55227 opened 2 months ago

55227 commented 2 months ago

Thank you for your contribution! Could you please release the datasets you used or provide more details about how they were constructed?

shunliu01 commented 2 months ago

have you successfully run the training pipeline till now??

55227 commented 2 months ago

谢谢回复,已经运行成功了,下载了Weakly-Supervised-Camouflaged-Object-Detection-with-Scribble-Annotations提供的训练集和测试集,但是不知道是否与你们使用的数据集相一致??(如果可以,还是想要你们数据集部分的实现细节)但是在运行过程中,我又遇到了别的问题,希望您能解答 ①readme.md中提供的命令行参数中指定trainsize为36,这与SoftPositionEmbed类中build_grid的传入参数resolution相矛盾,论文中提及所有训练集和测试集图像尺寸为352,但这也同样矛盾,导致报错 ②代码中没有论文中所提及的SAM产生伪标签的部分,但这一部分占据了论文50%以上的篇幅,我对这一部分很感兴趣,请问是否能够开源这一部分的代码呢 期待您的解答,谢谢!