search

meteorshowers / RCF-pytorch

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

adddilation-stridewith8 #3

Closed shiyanrubing closed 5 years ago

shiyanrubing commented 5 years ago

您好，我最近在学习您的RCF-pytorch 代码， 遇到了一点问题：adddilation-stridewith8 是什么意思？ 是要将涉及到dilation卷积的步长调整到8？ 比如， 在您的 models.py 文件中定义的类 class FlipConv2d(_ConvNd): def init(self, in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True): 是要将 stride=1 调整为 8。 不知我的理解是否正确？ 提前多谢您的指导哈，祝您有一个愉快的中秋。

meteorshowers commented 5 years ago

sorry for the late response~ this code is not for rcf model

shiyanrubing commented 5 years ago

where or which are the parameters of RCF model?

meteorshowers commented 5 years ago

just load vgg model is ok

shiyanrubing commented 5 years ago

I mean that the parameters are not prepared model parameters, and they are the ones like adam initial learning rating, weight decay and so on.

shiyanrubing commented 5 years ago

In your "Method", such as adam1e-4、 sgd1e-6-tunelr-caffecrop-fixbalanceweightto1.1, not for RCF model?

meteorshowers commented 5 years ago

because pytorch is different from caffe, so, I change the optimizer to adam. the caffe version use the sgd optimizer.