用于突出物体检测的深度对比度学习

由于使用深度卷积神经网络（CNN）提取的强大特征，突出物体检测最近取得了实质性进展。然而，现有的基于CNN的方法是在斑块级而不是像素级进行操作。由此产生的显著性地图通常是模糊的，尤其是在显著性物体的边界附近。此外，图像斑块被视为独立的样本，即使它们是重叠的，也会在计算和存储中产生大量的冗余。在本文中，我们提出了一个端到端的深度对比网络来克服上述的限制。我们的深度网络由两个互补的部分组成，一个是像素级的完全卷积流，一个是分段的空间汇集流。第一个流直接从输入图像中产生一个具有像素级精度的显著性地图。第二个数据流非常有效地提取分段特征，并更好地模拟物体边界上的显著性不连续现象。最后，一个完全连接的CRF模型可以被选择性地纳入，以改善这两个流的融合结果中的空间一致性和轮廓定位。实验结果表明，我们的深度模型明显改善了技术水平。

Two streams of our deep contrast network

Results

Visual comparison of saliency maps generated from state-of-the-art methods, including our DCL and DCL⁺. The ground truth (GT) is shown in the last column. DCL⁺ consistently produces saliency maps closest to the ground truth.

Quantitative Comparison

Comparison of precision-recall curves of 11 saliency detection methods on 3 datasets. Our MDF, DCL and DCL⁺ (DCL with CRF) consistently outperform other methods across all the testing datasets.

Comparison of precision, recall and F-measure (computed using a per-image adaptive threshold) among 11 different methods on 3 datasets.

Comparison of quantitative results including maximum F-measure (larger is better) and MAE (smaller is better). The best three results are shown in red, blue, and green , respectively.