meteorshowers / RCF-pytorch

Richer Convolutional Features for Edge Detection model in pytorch CVPR2017
509 stars 145 forks source link

不能复现结果 #41

Closed sunshine-ssl closed 4 years ago

sunshine-ssl commented 4 years ago

您好,我用扩充过的BSDS500数据集训练,ODS最高到0.778,请问您训练的时候最大迭代次数和stepsize设置的是多少呢,我用BSDS500加PASCAL数据集训练的时候出来的是全灰的图,不知道是哪里有问题

huangwefeng121 commented 4 years ago

您好,我用扩充过的BSDS500数据集训练,ODS最高到0.778,请问您训练的时候最大迭代次数和stepsize设置的是多少呢,我用BSDS500加PASCAL数据集训练的时候出来的是全灰的图,不知道是哪里有问题

请问怎么计算ODS和F-measure?

StormArcher commented 4 years ago

你们有没有用作者的模型直接测试?

meteorshowers commented 4 years ago

@sunshine-ssl you can change the learning rate to fit the dataset to stable the training process