DeepCO3: Deep Instance Co-Segmentation by Co-Peak Search and Co-Saliency Detection

Hsu, Ken-Yuh; Lin, Yen‐Yu; Chuang, Yung-Yu

doi:10.1109/cvpr.2019.00905

Cited by 66 publications

(26 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently deep-based models simultaneously explore the intra-and inter-image consistency in a supervised manner with different approaches, such as graph convolution networks (GCN) [32,33,34], self-learning methods [15,35], inter-image co-attention with PCA projection [1] or recurrent units [36], correlation techniques [37], quality measurement [38], or co-clustering [39]. Some methods exploit multi-task learning to simultaneously optimize the co-saliency detection and co-segmentation [40] or co-peak search [6]. Other works explore hierachical features from multi-scale [41], multi-stage [42], or multi-layer [43] features.…”

Section: Related Workmentioning

confidence: 99%

“…That is, both intra-class compactness and inter-class separability should be simultaneously maximized. With this favorable feature CoSOD is thus often employed as a pre-processing step for various computer vision tasks, such as image retrieval [2], image quality assessment [3], collection-based crops [4], co-segmentation [5,6], semantic segmentation [7], image surveillance [8], video analysis [9], video co-localization [10], etc.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Group Collaborative Learning for Co-Salient Object Detection

Fan¹,

Fan²,

Fu³

et al. 2021

Preprint

View full text Add to dashboard Cite

We present a novel group collaborative learning framework (GCoNet) capable of detecting co-salient objects in real time (16ms), by simultaneously mining consensus representations at group level based on the two necessary criteria: 1) intra-group compactness to better formulate the consistency among co-salient objects by capturing their inherent shared attributes using our novel group affinity module; 2) inter-group separability to effectively suppress the influence of noisy objects on the output by introducing our new group collaborating module conditioning the inconsistent consensus. To learn a better embedding space without extra computational overhead, we explicitly employ auxiliary classification supervision. Extensive experiments on three challenging benchmarks, i.e., CoCA, CoSOD3k, and Cosal2015, demonstrate that our simple GCoNet outperforms 10 cutting-edge models and achieves the new state-ofthe-art. We demonstrate this paper's new technical contributions on a number of important downstream computer vision applications including content aware co-segmentation, co-localization based automatic thumbnails, etc.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Group Collaborative Learning for Co-Salient Object Detection

Fan¹,

Fan²,

Fu³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…CosegCA provided an efficient instant group cosegmentation method to reduce complexity for co-segmenting several images (more than 2) through group average channelwise attention in linear time complexity. Hsu et al presented instance co-segmentation aiming to recognize and segment all instances belonging to the same class from two images in [42]. They leveraged co-peak search and instance mask segmentation to outperform the state-of-the-art methods.…”

Section: Object Co-segmentationmentioning

confidence: 99%

Channel Transformer Network

2020

View full text Add to dashboard Cite

“…During training, we define a novel loss function on pairs of pixels such that it encourages the pixel affinities to align with the predicted segmentation probabilities. A few methods have also proposed using feature affinity as an explicit measure to improve segmentation in fully-supervised and co-segmentation settings [5,18,22,31,33]. Affinities have also been used in some weakly supervised approaches [1,2], but they are trained in a fully-supervised manner using pseudo ground truth derived from class activation maps, as opposed to our embeddings which are trained by minimizing disagreements between them and estimated segmentation output (Sect.…”

Section: Introductionmentioning

confidence: 99%