A5100427461- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Advanced Image Processing Techniques
- Image Enhancement Techniques
- Video Surveillance and Tracking Methods
- Anomaly Detection Techniques and Applications
- Advanced Image Fusion Techniques
- Robotics and Sensor-Based Localization
- Remote-Sensing Image Classification
- Generative Adversarial Networks and Image Synthesis
- Advanced Memory and Neural Computing
- Outsourcing and Supply Chain Management
- Industrial Vision Systems and Defect Detection
- Domain Adaptation and Few-Shot Learning
- Image and Signal Denoising Methods
- Image Processing and 3D Reconstruction
- Handwritten Text Recognition Techniques
- Image and Object Detection Techniques
- CCD and CMOS Imaging Sensors
- Collaboration in agile enterprises
- Multimodal Machine Learning Applications
- Advanced Vision and Imaging
Institute of Computing Technology
2022-2025
Chinese Academy of Sciences
2023-2025
University of Chinese Academy of Sciences
2022-2023
Fuzhou University
2023
Jiaxing University
2023
Harbin Institute of Technology
2020
Limited data usually cause deep neural networks to hold poor performance after training, and many generative models are proposed synthesize improve the of models. However, existing ignore capturing small defect details (e.g., features locations), resulting in that most cannot augment Defect Location Sensitive Data (DLS data) which ratio object size image is 20%) locations defects only on object. In this paper, we propose a new augmentation model, named GAN (DLS-GAN), address DLS problem....
This report introduces two high-quality datasets Flickr360 and ODV360 for omnidirectional image video super-resolution, respectively, reports the NTIRE 2023 challenge on 360° super-resolution. Unlike ordinary 2D images/videos with a narrow field of view, can represent whole scene from all directions in one shot. There exists large gap between image/video both degradation restoration processes. The is held to facilitate development super-resolution by considering their special...
FPN is a common component used in object detectors, it supplements multi-scale information by adjacent level features interpolation and summation. However, due to the existence of nonlinear operations convolutional layers with different output dimensions, relationship between levels much more complex, pixel-wise summation not an efficient approach. In this paper, we first analyze design defects from pixel feature map level. Then, novel parameter-free pyramid networks named Dual Refinement...
Object detection on panoramic/spherical images has been developed rapidly in the past few years, where IoU-calculator is a fundamental part of various detector components, i.e. Label Assignment, Loss and NMS. Due to low efficiency non-differentiability spherical Unbiased IoU, approximate IoU methods have proposed recently. We find that key these map boxes planar boxes. However, there exists two problems methods: (1) they do not eliminate influence panoramic image distortion; (2) break...
Recently, many semi-supervised object detection (SSOD) methods adopt teacher-student framework and have achieved state-of-the-art results. However, the teacher network is tightly coupled with student since an exponential moving average (EMA) of student, which causes a performance bottleneck. To address coupling problem, we propose Cycle Self-Training (CST) for SSOD, consists two teachers T1 T2, students S1 S2. Based on these networks, cycle self-training mechanism built, i.e., S1$\rightarrow...
<title>Abstract</title> The importance of the attention mechanism in CV is growing, as it allows a neural network to focus more on what should pay to. Channel and spatial are two basic strategies now use. Using one them alone can enhance some level, while combining beneficial, but adds computational burden. We propose Spatial Attention Fusion Module(CSAFM), besides use channel information, GroupNorm reorganization operation applied ability feature extraction representation maps, which...
Displaying high-quality images on edge devices, such as augmented reality is essential for enhancing the user experience. However, these devices often face power consumption and computing resource limitations, making it challenging to apply many deep learning-based image compression algorithms in this field. Implicit Neural Representation (INR) an emerging technology that offers two key benefits compared cutting-edge autoencoder models: low computational complexity parameter-free decoding....
Strategic alliance provides a new way for logistics enterprises to increase competitiveness and adapt the competitive environment based on win-win. This paper defines strategic in analyses its causes forming manifestation. A reasonable profit distribution is key ensuring success of alliance, determine ratio. According principle that all an should have equal responsibilities, rights, interests, risks, this constructs model proves feasibility superiority model. case also given show how applied...
In order to reduce the time consuming and expensive process of manually annotating data, achieve purpose lightweight deployment. this paper, an object detection method for weakly supervised learning with discrimination mechanism is proposed. We introduce classification branch location based on Darknet-53 backbone network YOLO model, utilize Global Average Pooling (GAP) Softmax complete selected areas, adopt activation map location. addition, we use a model compression pruning operations,...
Recent advances of Transformers have brought new trust to computer vision tasks. However, on small dataset, is hard train and has lower performance than convolutional neural networks. We make transformers as data-efficient networks by introducing multi-focal attention bias. Inspired the distance in a well-trained ViT, we constrain self-attention ViT multi-scale localized receptive field. The size field adaptable during training so that optimal configuration can be learned. provide empirical...