yxqAIxp

Data Enhancement

1

Using a random cropping method, what is the difference between training and testing？

can I get more details for RFConv2d? Thanks.

4

from rfconv import RFConv2d self.conv = RFConv2d(in_channels, channels*radix, kernel_size, stride, padding, dilation, groups=groups*radix, bias=bias, average_mode=rectify_avg, **kwargs) What is rfconv?

眼睛闭眼的时候 眼睛部分是没有区分开 这是为什么 不考虑眼睛的位置嘛

landmark返回原图的坐标有误！代码应该改成： # landmark of (5L, 2L) from [0,1] to real range def reprojectLandmark(self, landmark, dx, dy, edx, edy): landmark_= np.asarray(np.zeros(landmark.shape)) for i, point in enumerate(landmark): x = point[0] * (self.w+dx+edx)...

我这里有个小疑问，人脸特征点训练的过程：人脸区域是按照人脸特征点的标签去截取的，所以训练一般会比较好，但如果测试则第一步需要人脸检测，请问人脸检测训练的人脸框是不是也得按照人脸特征点的标签去截取呢？要不然人脸框不一致，有的效果好，有的效果则很差？

特征点增加至68点训练，效果很不好

2

大神，你好，请问我想人脸和特征点一起训练，特征点是68点（300W），但是训练出来的效果不是很好，我看你五点的效果还是很好的，请问我是不是要改变训练的方法？  

Difference between isnet.pth and isnet-general-use.pth

1

Thanks for your great work. what are the training differences between isnet.pth and isnet-general-use.pth, as isnet-general-use works much better than isnet.pth.

Final value of training loss

1

Thanks for your great work! What is the final value of training loss that can be achieved by using intermediate supervision loss？