search

yun-liu / RCF

Richer Convolutional Features for Edge Detection
Other
761 stars 259 forks source link

One image has many ground truths #100

Open TeerathChandani opened 5 years ago

TeerathChandani commented 5 years ago

Currently, I am facing an issue in edge detection, that one image has many ground truths (one image with many edge detected). Suppose image img ---> {c1,c2,c3,c4,c5}, having five ground truth, these are edge detected image. So how did you train algorithm with many ground truths? You considered pairs like img-->c1 img-->c2 img-->c3 img-->c4 img-->c5

Currently, I have download data, but not sure how I should this issue. I read your paper RCF.

https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/resources.html,

yun-liu commented 5 years ago

In my paper, I have addressed this problem in the loss function. Specifically, I average all ground truth for each image and binarize this averaged edge map using a threshold. Please see section 3.2 in the RCF paper for more details.

TeerathChandani commented 5 years ago

Thanks. But threshold exact value is not mentioned?

yun-liu commented 5 years ago

I have provided the threshold values in may paper. Please read the section of loss function and experiment part.