Cascaded Partial Decoder for Fast and Accurate Salient Object Detection

By Zhe Wu, Li Su, Qingming Huang. Computer Vision and Pattern Recognition, 2019

953 citations recorded by the source corpus.

Existing state-of-the-art salient object detection networks rely on aggregating multi-level features of pre-trained convolutional neural networks (CNNs). However, compared to high-level features, low-level features contribute less to performance. Meanwhile, they raise more computational cost because of their larger spatial resolutions. In this paper, we propose a novel Cascaded Partial Decoder (CPD) framework for fast and accurate salient object detection. On the one hand, the framework constructs partial decoder which discards larger resolution features of shallow layers for acceleration. On the other hand, we observe that integrating features of deep layers will obtain relatively precise saliency map. Therefore we directly utilize generated saliency map to recurrently optimize features of deep layers. This strategy efficiently suppresses distractors in the features and significantly improves their representation ability. Experiments conducted on five benchmark datasets exhibit that the proposed model not only achieves state-of-the-art but also runs much faster than existing models. Besides, we apply the proposed framework to optimize existing multi-level feature aggregation models and significantly improve their efficiency and accuracy.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/66ce56946214ab496c260012d828f7883ebfca22

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Global Context-Aware Progressive Aggregation Network for Salient Object Detection (2020, background): “…(Li et al. 2018), RANet (Chen et al. 2018), PAGR (Zhang et al. 2018b), PiCANet-R (Liu, Han, and Yang 2018), DGRL (Wang et al. 2018), R 3 Net (Deng et al. 2018), BMPM (Zhang et al. 2018a), RADF (Hu et al. 2018), CPD-R (Wu, Su, and Huang 2019), BASNet (Qin et al. 2019) and Pool-Net (Liu et al. 2019).”
Recent advances and clinical applications of deep learning in medical image analysis (2021, background): “Aside from the semi-supervised learning strategy, there are three critical components in the model responsible for the good performance: parallel partial decoder (PPD) (Wu et al., 2019a), reverse attention (RA) (Chen et al., 2018b), and edge attention (Zhang et al., 2019).”
Polyp-PVT: Polyp Segmentation with Pyramid Vision Transformers (2021, methodology): “To balance the accuracy and computational resources, we follow recent popular practices [5], [75] to implement the cascaded fuse module (CFM).”
CaraNet: context axial reverse attention network for segmentation of small medical objects (2021, methodology): “Experiments have shown that, low-level features are more computationally costly but contribute less to improve segmentation performance [5].”
Kubric: A scalable dataset generator (2022, methodology): “Classical methods involve active-contour [47, 69] or graphcut [20, 104] techniques, but there also exist techniques with human-in-the-loop [72, 81, 100], and more recently deep learning variants [30, 73, 74, 105, 117].”
Suppress and Balance: A Simple Gated Network for Salient Object Detection (2020, methodology): “And many methods aim at combining multilevel features in either the encoder [69,53,67,57,37,59] or the decoder [29,20,70,59].”
Context-aware Cross-level Fusion Network for Camouflaged Object Detection (2021, background): “…UNet++ [Zhou et al., 2018], PiCANet [Liu et al., 2018], MSRCNN [Huang et al., 2019a], BASNet [Qin et al., 2019], PFANet [Zhao and Wu, 2019], CPD [Wu et al., 2019], HTC [Chen et al., 2019], EGNet [Zhao et al., 2019], ANetSRM [Le et al., 2019], SINet [Fan et al., 2021a] and MirrorNet [Yan et al.,…”
Camouflaged Object Detection (2020, methodology): “GOD vs. SOD Baselines.”
UACANet: Uncertainty Augmented Context Attention for Polyp Segmentation (2021, background): “They added parallel partial decoder from [28] to combine low-level feature maps from the backbone network because the architecture from [5] obviously lack of feature sharing for both top-down stream and bottom-up stream.”
Multi-Scale Interactive Network for Salient Object Detection (2020, methodology): “30% over the second-best method CPD [41] with the VGG-16 as the backbone and EGNet [56] with the ResNet50 as the backbone, respectively.”
PraNet: Parallel Reverse Attention Network for Polyp Segmentation (2020, methodology): “[29], compared with high-level features, low-level features demand more computational resources due to their larger spatial resolutions, but contribute less to performance.”
Visual Saliency Transformer (2021, methodology): “For RGB SOD, we compare our VST with 12 state-of-theart RGB SOD models, including GateNet [84], CSF [20], LDF [69], MINet [49], ITSD [87], EGNet [82], TSPOANet [41], AFNet [17], PoolNet [35], CPD [70], BASNet [55], and PiCANet [37].”
Defocus Blur Detection via Depth Distillation (2020, methodology): “methods on some relevant tasks: such as bidirectional feature pyramid network with recurrent attention (BDRAR [47]) and direction-aware attention (DSC [13]) for shadow detection, boundary-aware loss (BAS [30]) and cascaded partial decoder (CPD [37]) for salient object detection.”
A Survey of Deep Learning-Based Object Detection (2019, background): “[232] present that discarding the shallower layer features can achieve fast speed and the deeper layer features are sufficient to obtain precisely salient map.”
F3Net: Fusion, Feedback and Focus for Salient Object Detection (2019, methodology): “…will extract its features at ﬁve levels, denoted as Because low level features bring too much computational cost but little performance improvement (Wu, Su, and Huang 2019), we only use features of the last four levels f 2 , f 3 , f 4 , f 5 , which have lower resolutions and cost less computation.”
Concealed Object Detection (2021, methodology): “13, each TEM component includes four parallel residual branches { b i , i = 1 , 2 , 3 , 4 } with different dilation rates d ∈ { 1 , 3 , 5 , 7 } and a shortcut branch (gray arrow), respectively.”
Camouflaged Object Detection with Feature Decomposition and Edge Reconstruction (2023, neutral): “CHAMELEON (76 images) CAMO (250 images) COD10K (2,026 images) NC4K (4,121 images) Methods Publications Backbones M ↓ Fβ ↑ Eφ ↑ Sα ↑ M ↓ Fβ ↑ Eφ ↑ Sα ↑ M ↓ Fβ ↑ Eφ ↑ Sα ↑ M ↓ Fβ ↑ Eφ ↑ Sα ↑ Common Setting: Single Input Scale and Single Stage CPD [47] CVPR19 ResNet50 0.”