Question about the RCF edge detection framework.

[rrryan2016]

Thanks for your great work and kind sharing.

I successfully reproduce the edge detection result, as 2018.jpeg: 2018.png: After NMS process, it turns like:

However, when it switch the normal convolution to some special convolution in the same backbone(VGG16), the result is kinda wacky:

After same NMS process of edge thinning, it turns like more coarser in detail when compared with result above :

Also, after evaluation, the ODS-F is 0.685891 and OIS-F is 0.703985, both are much less than the paper, 0.788864 and 0.806692.

Could any one please point out where is the potential problem lie in, cause I am really confused about it, and have no much experience in edge detection.

Thanks in advance !

[rrryan2016]

I guess the problem may be that my backbone is not well trained, as it could extract detailed features as shown in the illustration pictures.

The pretrained on part of ImageNet is as below, and I am still working at it.

	Baseline Backbone	My Backbone
Train TOP1 Acc.	83.792	72.572
Train TOP5 Acc.	94.957	89.900
Val TOP1 Acc.	82.760	74.800
Val TOP5 Acc.	94.580	90.760

[rrryan2016]

I well-trained my backbone again, as below:

	Baseline Baseline	My backbone
Train TOP1 Acc.	83.792	81.660
Train TOP5 Acc.	94.957	93.940
Val TOP1 Acc.	82.760	81.280
Val TOP5 Acc.	94.580	94.100

However, the wacky problem still occurs,

Could anyone give me any advice? Thanks in advance.