A5102938165- Advanced Neural Network Applications
- Anomaly Detection Techniques and Applications
- Video Surveillance and Tracking Methods
- Advanced Wireless Communication Techniques
- Advanced Vision and Imaging
- Error Correcting Code Techniques
- Human Pose and Action Recognition
- Robotics and Sensor-Based Localization
- Underwater Vehicles and Communication Systems
- Algorithms and Data Compression
- 3D Shape Modeling and Analysis
- Optical measurement and interference techniques
- Cooperative Communication and Network Coding
- Robot Manipulation and Learning
- Indoor and Outdoor Localization Technologies
- Advanced Wireless Communication Technologies
- Caching and Content Delivery
- Telecommunications and Broadcasting Technologies
- Advanced Graph Neural Networks
Shanghai Jiao Tong University
2019-2024
Pedestrian detection in a crowd is challenging task due to high number of mutually-occluding human instances, which brings ambiguity and optimization difficulties the current IoU-based ground truth assignment procedure classical object methods. In this paper, we develop unique perspective pedestrian as variational inference problem. We formulate novel efficient algorithm for by modeling dense proposals latent variable while proposing customized Auto-Encoding Variational Bayes (AEVB)...
The min-sum (MS) algorithm can decode Low-density parity-check (LDPC) codes with low computational complexity at the cost of slight performance loss. It is an effective way to realize hardware implementation decoder by quantizing floating belief messages (i.e., check-to-variable and variable-to-check messages) into low-resolution 2–4 bits) versions. However, such a lead severe degradation due finite precision effect. In this paper, we propose neural-network optimized decoding (NOLD) for LDPC...
Compared with the sum-product algorithm (SPA) high decoding complexity for low-density parity-check (LDPC) codes, its approximated version, min-sum (MSA), reduces computational at cost of slight performance degradation. In order to compensate oversized check-node messages in MSA, effect clipping on variable-node is investigated under two-bit resolution. Our results show that clipped MSA degrades bit error rate (BER) and increases low BER as threshold enlarges. Based these results, we propose...
We present a novel method for 3D human body reconstruction with multi-view images from calibration-free cameras by fusion explicit visibility modelling. Existing methods usually establish geometric constraints using accurate camera intrinsic and extrinsic parameters. Despite remarkable performances, calibration often requires complex operations additional maintenance to fix positions angles, which restrict its applicability real-world scenarios. In contrast, we leverage vertex-wise...
The intelligent reflecting surface (IRS) has been recently brought up and foreseen to be one of the revolutionary technologies for future communication systems. A significant number low-cost passive elements (PAs) can effectively reflect incident signals collaboratively achieve beamforming by controlling phase shifts at PAs. In this paper, we study problem (PBP) in IRS-assisted multiple-input single-output (MISO) wireless system with imperfect channel station information (CSI). Considering...
Estimating 6D pose of the object from a single image is es-sential for robotic manipulation. Many recent learning-based methods directly regress 2D-3D points corre-spondence. The problem that, these only make use visible information single-view image, resulting ambiguity network to solve limited cor-responding pairs. To overcome this problem, paper intro-duces INVNet, integrating invisible into visi-ble correspondence model geometry features 3D object. Instead reconstruct coordinate...
The high computational cost and huge amount of parameters 3D Convolutional Neural Networks(CNN) limit the deployment CNN-based models to lightweight devices with weak computing ability. Applying Winograd layer that combines idea pruning algorithms convolution is a promising solution. However, irregular unstructured sparsity makes it difficult implement efficient compression. In this paper, we propose Regular Mask Pruning obtain regular sparse can be easily compressed. Furthermore, present...
It has been witnessed that the learned data compression techniques outperformed conventional ones. However, non-deterministic floating-point calculation makes probability prediction inconsistent between sender and receiver, disabling practical applications. We propose to use integer network relieve this problem focus on graph lossless compression. Firstly, we an adaptive fixed-point format, AdaFixedPoint, which can convert a model, convolution layers one with minimal precision loss enable...
Pedestrian detection in a crowd is challenging task due to high number of mutually-occluding human instances, which brings ambiguity and optimization difficulties the current IoU-based ground truth assignment procedure classical object methods. In this paper, we develop unique perspective pedestrian as variational inference problem. We formulate novel efficient algorithm for by modeling dense proposals latent variable while proposing customized Auto Encoding Variational Bayes (AEVB)...