Boundary-Guided Camouflaged Object Detection

Abstract

Forecasting the future trajectories of multiple agents is a core technology for human-robot interaction systems. To predict multi-agent trajectories more accurately, it is inevitable that models need to improve interpretability and reduce redundancy. However, many methods adopt implicit weight calculation or black-box networks to learn the semantic interaction of agents, which obviously lack enough interpretation. In addition, most of the existing works model the relation among all agents in a one-to-one manner, which might lead to irrational trajectory predictions due to its redundancy and noise. To address the above issues, we present Hypertron, a human-understandable and lightweight hypergraph-based multi-agent forecasting framework, to explicitly estimate the motions of multiple agents and generate reasonable trajectories. The framework explicitly interacts among multiple agents and learns their latent intentions by our coarse-to-fine hypergraph convolution interaction module. Our experiments on several challenging real-world trajectory forecasting datasets show that Hypertron outperforms a wide array of state-of-the-art methods while saving over 60% parameters and reducing 30% inference time.

Horizontal Feature Pyramid Network for Object Detection in UAV Images

“…To demonstrate the effectiveness of the proposed method, we compare it with 24 state-of-the-art methods, including 10 salient object detection methods (i.e., BAS-Net [33], CPD-R [45], EGNet [54], SCRN [46], F 3 Net [44], CSNet [13], SSAL [50], ITSD [58], UCNet [49], and VST [22]), and 14 COD methods (i.e., SINet [11], SLSR [25], PFNet [28], MGL-R [48], UJSC [19], PreyNet [51], BSA-Net [59], C 2 FNet [36], UGTR [47], OCE-Net [21], BGNet [37], SegMaR [16], ZoomNet [30], and SINet-v2 [10]). All the predictions of competitors are either provided by the authors or generated by models retrained based on opensource codes.…”

“…Besides, the multi-task learning framework is commonly used for COD. These methods generally introduce auxiliary tasks such as classification [18], edge/boundary detection [37,48,59], and object ranking [25]. Furthermore, some methods detect camouflaged objects by mimicking behavior patterns or visual mechanics of predators such as the search and identification process [10], and zooming in…”

“…We show more quantitative experimental results on three benchmark COD datasets. The methods used in the experiments for comparison include 10 SOD methods (BAS-Net [33], CPD [45], EGNet [54], SCRN [46], F 3 Net [44], CSNet [13], SSAL [50], ITSD [58], UCNet [49], VST [22]) and 13 COD methods (SINet [11], SLSR [25], PFNet [28], MGL-R [48], UJSC [19], C 2 FNet [36], UGTR [47], BSA-Net [59], OCE-Net [21], BGNet [37], SegMaR [16], Zoom-Net [30], SINet-v2 [10]).…”

Boundary-guided network for camouflaged object detection

“…Recently, with the advancement of large-scale benchmark datasets [9]- [11], a plethora of COD [8], [10], [12]- [14] methods have been proposed and have exhibited exceptional performance in diverse and intricate scenarios. In general, the detection process of these models can be summarized into , FAPNet [7] and PreyNet [8], our method preserves better details and is competent to distinguish the entire target from the background.…”

“…Despite the effectiveness of this widely adopted framework in addressing various challenging scenarios, there exist certain overlooked flaws that significantly hinder its overall performance. Concretely, in the first step, the majority of COD models employ low-resolution images (e.g., 352 × 352 [7], [15], [16], 416 × 416 [14]) as their input. However, camouflaged objects can be small in size and exhibit fuzzy boundaries.…”