Awesome
TrasformerSOD
Difficulty-aware Salient Object Detection via Deeply-supervised Transformer.
A unified framework for Trasformer supervised/weakly supervised SOD, RGBD SOD, COD.
Code will be released very soon.
Our results
The saliency maps of DUTS, ECSSD, DUT, HKU-IS, PASCAL-S, SOD dataset in fully supervised manner are as: saliency maps1
S_measure | F_measure | E_measure | MAE | |
---|---|---|---|---|
DUTS | 0.91725 | 0.89585 | 0.95016 | 0.02462 |
ECSSD | 0.94301 | 0.94535 | 0.96665 | 0.02246 |
DUT | 0.86163 | 0.80710 | 0.89227 | 0.04819 |
HKU-IS | 0.93410 | 0.93195 | 0.96689 | 0.02154 |
PASCAL | 0.88257 | 0.88097 | 0.92259 | 0.05063 |
SOD | 0.86088 | 0.86524 | 0.89719 | 0.06080 |
The saliency maps of NJU2K,STERE,DES,NLPR,LFSD,SIP dataset in fully supervised manner are as: saliency maps2
S_measure | F_measure | E_measure | MAE | |
---|---|---|---|---|
NJU2K | 0.92714 | 0.92206 | 0.95570 | 0.02777 |
STERE | 0.91871 | 0.90235 | 0.95054 | 0.03114 |
DES | 0.93886 | 0.93149 | 0.97045 | 0.01541 |
NLPR | 0.93473 | 0.91704 | 0.96420 | 0.01941 |
LFSD | 0.87720 | 0.86676 | 0.90660 | 0.05896 |
SIP | 0.90487 | 0.91094 | 0.94262 | 0.03602 |
The saliency maps of DUTS, ECSSD, DUT, HKU-IS, PASCAL-S, SOD dataset in weakly supervised manner are as: saliency maps3
S_measure | F_measure | E_measure | MAE | |
---|---|---|---|---|
DUTS | 0.860 | 0.823 | 0.915 | 0.040 |
ECSSD | 0.906 | 0.913 | 0.951 | 0.038 |
DUT | 0.838 | 0.768 | 0.888 | 0.056 |
HKU-IS | 0.899 | 0.893 | 0.954 | 0.034 |
PASCAL | 0.848 | 0.823 | 0.902 | 0.065 |
SOD | 0.817 | 0.818 | 0.872 | 0.080 |
The camouflage maps of CAMO, CHAMELEON, COD10K, NC4K dataset in weakly supervised manner are as: camouflage maps
S_measure | F_measure | E_measure | MAE | |
---|---|---|---|---|
CAMO | 0.86132 | 0.84558 | 0.92487 | 0.04681 |
CHAMELEON | 0.89951 | 0.86423 | 0.95401 | 0.02289 |
COD10K | 0.85013 | 0.78902 | 0.92432 | 0.02558 |
NC4K | 0.87993 | 0.85659 | 0.93243 | 0.03346 |
Contact
Please drop me an email for further problems or discussion: maoyuxin@mail.nwpu.edu.cn