| name | caffemodel | caffemodel_url | sha1 | gist_id |
|---|---|---|---|---|
Holistically-Nested Edge Detection on BSDS500 |
hed_pretrained_bsds.caffemodel |
2c5d7842f25f880eec62fc610b500c5cf2aa351d |
c6bd432f7347548b0187 |
This is a model from the paper:
Holistically-Nested Edge Detection
Saining Xie, Zhuowen Tu
ICCV 2015
This model was trained for the BSDS500 edge detection.
The input is expected in BGR channel order, with the following per-channel mean subtracted:
B 104.00698793 G 116.66876762 R 122.67891434
We refactored our code base and adopt the handy python wrapper for in-network interpolation and training/testing in this release. Thanks BVLC for this.
Details of training/testing our algorithm are available at https://github.com/s9xie/hed.
This model obtains 0.790 ODS for the fusion-layer output on BSDS500.