F3Net: Fusion, Feedback and Focus for Salient Object Detection

By Junhang Wei, Shuhui Wang, Qingming Huang. AAAI Conference on Artificial Intelligence, 2019

862 citations recorded by the source corpus.

Most of existing salient object detection models have achieved great progress by aggregating multi-level features extracted from convolutional neural networks. However, because of the different receptive fields of different convolutional layers, there exists big differences between features generated by these layers. Common feature fusion strategies (addition or concatenation) ignore these differences and may cause suboptimal solutions. In this paper, we propose the F3Net to solve above problem, which mainly consists of cross feature module (CFM) and cascaded feedback decoder (CFD) trained by minimizing a new pixel position aware loss (PPA). Specifically, CFM aims to selectively aggregate multi-level features. Different from addition and concatenation, CFM adaptively selects complementary components from input features before fusion, which can effectively avoid introducing too much redundant information that may destroy the original features. Besides, CFD adopts a multi-stage feedback mechanism, where features closed to supervision will be introduced to the output of previous layers to supplement them and eliminate the differences between features. These refined features will go through multiple similar iterations before generating the final saliency maps. Furthermore, different from binary cross entropy, the proposed PPA loss doesn't treat pixels equally, which can synthesize the local structure information of a pixel to guide the network to focus more on local details. Hard pixels from boundaries or error-prone parts will be given more attention to emphasize their importance. F3Net is able to segment salient object regions accurately and provide clear local details. Comprehensive experiments on five benchmark datasets demonstrate that F3Net outperforms state-of-the-art approaches on six evaluation metrics. Code will be released at https://github.com/weijun88/F3Net.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/b86cf0116da79318b07062b1d3456711fdf808aa

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

In Media Technology

Citations in the corpus

UC-Net: Uncertainty Inspired RGB-D Saliency Detection via Conditional Variational Autoencoders (2020, background): “Object-level visual saliency detection involves separating the most conspicuous objects that attract humans from the background [27, 2, 55, 63, 38, 29, 62].”
Inf-Net: Automatic COVID-19 Lung Infection Segmentation From CT Images (2020, methodology): “nce. In addition, compared with the standard BCE loss, Lw BCE puts more emphasis on hard pixels rather than assigning all pixels equal weights. The deﬁnitions of these losses are the same as in [70], [71] and their effectiveness has been validated in the ﬁeld of salient object detection. Note that t”
Uncertainty-aware Joint Salient Object and Camouflaged Object Detection (2021, methodology): “SCRN [48], F(3)Net [46], ITSD [59], and it can indeed improve model performance.”
Polyp-PVT: Polyp Segmentation with Pyramid Vision Transformers (2021, background): “LIoU(·) and LBCE(·) are the weighted intersection over union (IoU) loss [83] and weighted binary cross entropy (BCE) loss [83], which restrict the prediction map in terms of the global structure (object-level) and local details (pixel-level) perspectives.”
Concealed Object Detection (2021, background): “The definitions of these losses are the same as in [59], [70] and their effectiveness has been validated in”
PraNet: Parallel Reverse Attention Network for Polyp Segmentation (2020, background): “The definitions of these losses are the same as in [21,26] and their effectiveness has been validated in the field of salient object detection.”
Simultaneously Localize, Segment and Rank the Camouflaged Objects (2021, methodology): “For the latter, we adopt the pixel position aware loss as in [49] to produce predictions with higher structure accuracy.”
Shallow Attention Network for Polyp Segmentation (2021, background): “Because they have higher resolutions and contain richer details, compared with deep ones [19].”
Salient Object Detection via Integrity Learning (2021, methodology): “5) FNR is the false negative ratio, which is computed by: FNðx; yÞ ¼ 1; Gðx; yÞ ¼ 1 & P ðx; yÞ ¼ 0; 0; others: (7) FNR ¼ PW x¼1 PH y¼1 FNðx; yÞPW x¼1 PH y¼1 Gðx; yÞ 100%; (8) TABLE 1 Quantitative Results on Six Datasets Summary ECSSD [76] PASCAL-S [78] DUTS [75] HKU-IS [77] OMRON [15] SOD [79] Metho”
BBS-Net: RGB-D Salient Object Detection with a Bifurcated Backbone Strategy Network (2020, neutral): “deep methods [52] usually leverage CNNs to extract multilevel multi-scale features from RGB images and then aggregate them to predict the ﬁnal saliency map. Such multilevel multi-scale features [53], [54] can help the model better understand the contextual and semantic information to generate high-q”
Zoom In and Out: A Mixed-scale Triplet Network for Camouflaged Object Detection (2022, methodology): “For a fair comparison on SOD, the proposed model is retrained on DUTS [44] following the training strategies and techniques of [35,46,47,51,64].”
Camouflaged Object Segmentation with Distraction Mining (2021, methodology): “…detection methods DSC [20] and BDRAR [72]; medical image segmentation methods UNet++ [71] and PraNet [13]; salient object detection methods PiCANet [33], BASNet [44], CPD [55], PFANet [69], EGNet [67], F3Net [53], GC-PANet [5], and MINet-R [40]; and camouﬂaged object segmentation method SINet [12].”
Automatic Polyp Segmentation via Multi-scale Subtraction Network (2021, methodology): “We use the same definitions as in [7,25,19] and their effectiveness has been validated in theses works.”