search

yun-liu / RCF

Richer Convolutional Features for Edge Detection
Other
760 stars 260 forks source link

what should I do after get results through rcf-singlescale-nyud？？ #45

Open theonels opened 6 years ago

theonels commented 6 years ago

There is a saying In your paper:

When evaluating, standard NMS is applied to thin detected deges

However, the note in this READ.ME point out that we should do NMS and edge thinning. I am quite confused now...Is there one step or two? Have the results I got from rcf-singlescale-nyud already been done NMS or NOT？I have got the tools about your link. It seems like another way to train a model and evaluate results. Thanks a lot!

yun-liu commented 6 years ago

Standard NMS is applied to thin detected deges, which means NMS is for thinning edges. (这句话的中文翻译：在评测时，用标准的NMS去细化检测出的边缘。) There is a NMS function in dollar's edge toolbox you can use.

theonels commented 6 years ago

uh...Here is what I mean. In your paper:

When evaluating, standard NMS is applied to thin detected deges

In the README.md of your github:

Note: Before evaluating the predicted edges, you should do the standard non-maximum suppression (NMS) and edge thinning.

You do another edge thinning after NMS?

yun-liu commented 6 years ago

No, only NMS is enough.

theonels commented 6 years ago

what's the tolerence you use when evaluate p/rF? Is it 0.0075 ?

yun-liu commented 6 years ago

As shown in the paper, the localization tolerance is 0.011 for NYUD dataset. This value is also used by other papers.

theonels commented 6 years ago

sorry for I missed it.

we increase localization tolerance from 0.0075 to 0.011, because images in NYUD dataset are larger than images in BSDS500 dataset.

As I know, localization tolerance = the number*the image diagonal. Here we set a larger number for a larger size image. Is it for get a higher F? or Will we loss some accuracy when deal with large size image? Thanks a lot!