search

pkuCactus / BDCN

The code for the CVPR2019 paper Bi-Directional Cascade Network for Perceptual Edge Detection
MIT License
341 stars 71 forks source link

Not as crisp results as shown in paper #56

Open Ton-lee opened 1 year ago

Ton-lee commented 1 year ago

Thanks to your excellent work very much but I was not able to reproduce as crisp results as shown in your original paper (like in Figure 7 ). Your pretrained model was taken and the results were with thick edges like the following picture shows and training couldn't help.

Could you please provide your new final model if any? Or maybe there were any changes in your codes?

The following are the results shown in your paper v.s. my result with your pretrained model.

image

100075

pkuCactus commented 1 year ago

Thanks for your attention to our work , nms process is performed to get the final results, which you can find it the edge tool box refer to https://github.com/pdollar/edges/tree/master/private.