Segment anything model (SAM) developed by Meta AI Research has recently attracted significant attention. Trained on a large segmentation dataset of over 1 billion masks, SAM is capable segmenting any object certain image. In the original work, authors turned to zero-short transfer tasks (like edge detection) for evaluating performance SAM. Recently, numerous works have attempted investigate in various scenarios recognize and segment objects. Moreover, projects emerged show versatility as...

The emotional response of robotics is crucial for promoting the socially intelligent level human–robot interaction (HRI). development machine learning has extensively stimulated research on recognition robots. Our focuses gaits, a type simple modality that stores series joint coordinates and easy humanoid robots to execute. However, limited amount investigates HRI systems based indicating an existing gap in human emotion gait robotic response. To address this challenge, we propose...

Aiming at the problem of low detection accuracy grasping algorithm based on RGB information as input, this paper proposes a CSP-ResNet to improve algorithm. that chessboard effect will be produced when transposed convolution restores image fractional variability, which affect prediction model, designs fusion nearest neighbor interpolation upsampling and restore resolution feature map. To alleviate checkerboard by convolution. Namely FCG-Net (Fuse CSPS Grasp Net). This method takes first...

The concern over data and model privacy in machine learning inference as a service (MLaaS) has led to the development of private (PI) techniques. However, existing PI frameworks, especially those designed for large models such vision transformers (ViT), suffer from high computational communication overheads caused by expensive multi-party computation (MPC) protocols. encrypted attention module that involves softmax operation contributes significantly this overhead. In work, we present family...

Grasping a diverse range of novel objects from dense clutter poses great challenge to robots because the occlusion among these objects. In this work, we propose Pyramid-Monozone Synergistic Policy (PMSGP) that enables cleverly avoid most occlusions during grasping. Specifically, initially construct Pyramid Se quencing (PSP) sequence each object in scene into pyramid structure. By isolating layer-by-layer, grasp candidates will focus on single layer grasp. Then, devise Monozone Sampling (MSP)...

The option framework has shown great promise by automatically extracting temporally-extended sub-tasks from a long-horizon task. Methods have been proposed for concurrently learning low-level intra-option policies and high-level selection policy. However, existing methods typically suffer two major challenges: ineffective exploration unstable updates. In this paper, we present novel stable off-policy approach that builds on the maximum entropy model to address these challenges. Our...

A new approach to digital image signatures is proposed. The proposed has shown be resistant several kinds of processing and the JPEG lossy compression. Moreover, can extracted from watermarked without resorting original image.

This paper studies the prevention of premature failures LED backlights used in mobile devices that are subject to different use conditions. is a vitally important topic for consumer device manufacturers as life expectancy two identical from same production line may vary substantially under operating environments and These differences not addressed by traditional reliability assessment methods documented many electronics handbooks. The outlines prognostics approach condition-based monitoring...

Abstract In order to effectively improve the detection accuracy of remote sensing images in airport areas, basing on representative deep network Faster R-CNN as object method, a deeper basic ResNet and feature fusion component FPN are used extract more robust distinguishing features, add new fully connected layer end combine softmax classifier 4 logistic regression classifiers for according inter-class correlation object. Experiments show that improvement original brings 7.7% mAP 76.6% mAP....

Population aging is a growing issue for many metropolitan cities. With the proven effectiveness of assistive technology elderly care, smart home healthcare system that utilizes would facilitate independent living senior citizens with cognitive impairment alone. The main objective proposed to provide patients point-of-care solution minimal user intervention while reducing demand public services.

This study focuses on the over-fitting problem in training process of deep convolutional neural network model and poor robustness when is applied an occlusion environment. We propose a unique data augmentation method, In-and-Out. First, information variance enhanced through dynamic local operation while maintaining overall geometric structure image; compared with global our method effectively alleviates overfitting significantly improves generalization ability model. Then removal operation,...

In contrast to the human vision that mainly depends on shape for recognizing objects, deep image recognition models are widely known be biased toward texture. Recently, Meta research team has released first foundation model segmentation, termed segment anything (SAM), which attracted significant attention. this work, we understand SAM from perspective of texture \textit{v.s.} shape. Different label-oriented tasks, is trained predict a mask covering object based promt. With said, it seems...

We revisit the relationship between attention mechanisms and large kernel ConvNets in visual transformers propose a new spatial named Large Kernel Convolutional Attention (LKCA). It simplifies operation by replacing it with single convolution. LKCA combines advantages of convolutional neural networks transformers, possessing receptive field, locality, parameter sharing. explained superiority from both convolution perspectives, providing equivalent code implementations for each view....

The detection accuracy and speed of grasp models on benchmarks are the focal points concern in robotic grasping community. Especially a collaborative robot setting, safety model is an essential aspect that cannot be overlooked. In this paper, we explore how to enhance autonomous vision-guided grasping. Specifically, propose simple yet practical Safety-optimized Strategy, which consists two parts. first part involves depth prioritization, optimizing sequence from top bottom based order...

Integrating the artificial intelligence vision system into robots has significantly enhanced adaptability of grasping, but are vulnerable to potential backdoor threats. Currently, majority attacks focused on image classification and limited unimodal information single-object digital scenarios. In this work, we make first endeavor realize attack multimodal vision-guided robot grasping within high-clutter Specifically, propose a novel method named Shortcut-enhanced Multimodal Backdoor Attack...

This paper presents an efficient deep reinforcement learning (DRL) framework for online 3D bin packing (3D-BPP). The 3D-BPP is NP-hard problem significant in logistics, warehousing, and transportation, involving the optimal arrangement of objects inside a bin. Traditional heuristic algorithms often fail to address dynamic physical constraints real-time scenarios. We introduce novel DRL that integrates reliable physics algorithm object rearrangement stable placement. Our experiment show...