BASNet
POOLNet
評測對比
實驗設計：
1 優化器、權重衰減：Adam optimizer with a weight decay of 5e-4
2 學習率：and an initial learning rate of 5e5 which is divided by 10 after 15 epochs.
3 迭代次數：24 epochs in total
4 數據增強：random horizontal flipping
5 測試訓練圖片大小：input images are kept unchanged
6 顯著目標loss：standard binary cross entropy loss for salient object detection
7 邊緣檢測loss: balanced binary cross entropy loss [40] for edge detection.
[40] HED: Saining Xie and Zhuowen Tu. Holistically-nested edge detection. In ICCV, pages 1395–1403, 2015. 6