Awesome
Salient Object Detection via Dynamic Scale Routing
Saliency Maps
We provide the saliency maps (Fetch Code: iirk) for comparisions, including DUTS-OMRON, DUTS-TE, ECSSD, HKU-IS, PASCAL-S. To obtain the same score with our paper, we recommend the evaluation code provided by Feng Mengyang.
| Backbone | # Params | #FLOPs | Saliency maps | Pre-trained model |
|---|---|---|---|---|
| DPNet-50 | 27.1M | 9.2G | maps (Fetch Code: iirk) | model (Fetch Code: 6unj) |
| DPNet-101 | 44.7M | 12.6G | maps (Fetch Code: izwv) | model (Fetch Code: x8h4) |
| DPNet-152 | 59.1M | 16G | maps (Fetch Code: xsx5) | model (Fetch Code: vh5j) |
We also provid the saliency maps (Fetch Code: ezc8) of SOTA models .
SOC Saliency Maps
In the paper, we compare DPNet with 12 methods on SOC test set (1200 images). The SOC saliency maps of previous methods is borrowed from SRCN project, including DSS、NLDF、SRM、Amulet、DGRL、BMPM、PiCANet-R、R3Net、C2S-Net、RANet、CPD、AFN、BASNet、PoolNet、SCRN、SIBA、EGNet、F3Net、GCPANet、MINet.
Here, we also share our SOC saliency maps (Fetch code:rnsm) for comparison. To obtain the same score with our paper, we recommend the evaluation code provided by Fan Dengping.
Acknowledgement
Our work is based on F3Net. We fully thank their open-sourced code.