search

backseason / PoolNet

Code for our CVPR 2019 paper "A Simple Pooling-Based Design for Real-Time Salient Object Detection"
MIT License
629 stars 153 forks source link

Can you provide the pre-computed saliency maps? #23

Closed ShuhanChen closed 4 years ago

ShuhanChen commented 5 years ago

Can you provide the pre-computed saliency maps, both vgg and resnet version?

backseason commented 5 years ago

We have provided the resnet-50 version over here: https://github.com/backseason/PoolNet#6-pre-trained-models-pre-computed-results-and-evaluation-results

ShuhanChen commented 5 years ago

Thanks, I found it. BTW, can you also provide the vgg version?

ShuhanChen commented 5 years ago

There are two images missing on DUT-OMRON: sun_akxddynsopjifavt.jpg and sun_barwtmnxkuxyttlj.jpg, can you check?

I also found the max F results on DUT-OMRON reported in the paper are different with other papers, such as PiCANet_resnet-50 is 0.803 in paper CPD CVPR19, which is lower than 0.820 as shown in your paper. I also used your provided saliency maps and SalMetric, and only got 0.805, which is also lower than 0.833. Can you tell me why? Thanks.

ZhangC2 commented 4 years ago

Are the provided the pre-computed saliency maps resnet-50 version? I think they are the VGG version. I evalute the results generated by the provided resnet50-model (w/o edge), and i cannot get the simliar results reported in the paper.