Deep learning-based Joint Source-Channel Coding(JSCC)is a crucial component in semantic communication,and recent research has made significant progress in adapting to different channels.In this paper,we propose a mult...Deep learning-based Joint Source-Channel Coding(JSCC)is a crucial component in semantic communication,and recent research has made significant progress in adapting to different channels.In this paper,we propose a multi-stage progressive technique called Deep learning based Progressive Joint Source-Channel Coding(DP-JSCC).This approach partitions the source into multiple stages and transmits the signals continuously.The receiver gradually enhances the quality of image reconstruction by progressively receiving the signals,offering greater flexibility compared to existing dynamic rate transmission methods.The model adopts a lightweight architectural design,where we introduce an efficient module called the Inverted Shuffle Attention Bottleneck(ISAB)and incorporate self-attention mechanisms in the encoding and decoding process to capture signal correlations and establish long-range dependencies.Additionally,we introduce the Progressive Focus Weight Allocation(PFWA)method to improve the image reconstruction capability in progressive transmission tasks.These design enhance the expressive capacity of the model.Simulation results demonstrate that DP-JSCC can flexibly adjust the transmission rate according to requirements without the need for retraining or deployment,enabling continuous optimization of signals at different rates.Furthermore,compared to stateof-the-art JSCC methods,DP-JSCC exhibits advantages in terms of computational complexity,parameter count,and reconstruction performance.展开更多
Capturing high-fidelity normals from single face images plays a core role in numerous computer vision and graphics applications.Though significant progress has been made in recent years,how to effectively and efficien...Capturing high-fidelity normals from single face images plays a core role in numerous computer vision and graphics applications.Though significant progress has been made in recent years,how to effectively and efficiently explore normal priors remains challenging.Most existing approaches depend on the development of intricate network architectures and complex calculations for in-the-wild face images.To overcome the above issue,we propose a simple yet effective cascaded neural network,called Cas-FNE,which progressively boosts the quality of predicted normals with marginal model parameters and computational cost.Meanwhile,it can mitigate the imbalance issue between training data and real-world face images due to the progressive refinement mechanism,and thus boost the generalization ability of the model.Specifically,in the training phase,our model relies solely on a small amount of labeled data.The earlier prediction serves as guidance for following refinement.In addition,our shared-parameter cascaded block employs a recurrent mechanism,allowing it to be applied multiple times for optimization without increasing network parameters.Quantitative and qualitative evaluations on benchmark datasets are conducted to show that our Cas-FNE can faithfully maintain facial details and reveal its superiority over state-of-the-artmethods.The code is available at https://github.com/AutoHDR/CasFNE.git.展开更多
The incremental radiosity method has been shown to be an efficienttechnique for providing global illumination in dynamic environments as it exploitstemporal coherence in object space. This paper presents an accelerate...The incremental radiosity method has been shown to be an efficienttechnique for providing global illumination in dynamic environments as it exploitstemporal coherence in object space. This paper presents an accelerated incremental radiosity algorithm, which is based on a dynamically followed partial matrix.This not only reduces the computation cost in determining incremental form-factorswhen the geometrical relationships between objects are constantly changing, but alsosimplifies the management of user interaction with comparatively little storage cost.展开更多
文摘Deep learning-based Joint Source-Channel Coding(JSCC)is a crucial component in semantic communication,and recent research has made significant progress in adapting to different channels.In this paper,we propose a multi-stage progressive technique called Deep learning based Progressive Joint Source-Channel Coding(DP-JSCC).This approach partitions the source into multiple stages and transmits the signals continuously.The receiver gradually enhances the quality of image reconstruction by progressively receiving the signals,offering greater flexibility compared to existing dynamic rate transmission methods.The model adopts a lightweight architectural design,where we introduce an efficient module called the Inverted Shuffle Attention Bottleneck(ISAB)and incorporate self-attention mechanisms in the encoding and decoding process to capture signal correlations and establish long-range dependencies.Additionally,we introduce the Progressive Focus Weight Allocation(PFWA)method to improve the image reconstruction capability in progressive transmission tasks.These design enhance the expressive capacity of the model.Simulation results demonstrate that DP-JSCC can flexibly adjust the transmission rate according to requirements without the need for retraining or deployment,enabling continuous optimization of signals at different rates.Furthermore,compared to stateof-the-art JSCC methods,DP-JSCC exhibits advantages in terms of computational complexity,parameter count,and reconstruction performance.
基金supported by the National Natural Science Foundation of China(62072327)。
文摘Capturing high-fidelity normals from single face images plays a core role in numerous computer vision and graphics applications.Though significant progress has been made in recent years,how to effectively and efficiently explore normal priors remains challenging.Most existing approaches depend on the development of intricate network architectures and complex calculations for in-the-wild face images.To overcome the above issue,we propose a simple yet effective cascaded neural network,called Cas-FNE,which progressively boosts the quality of predicted normals with marginal model parameters and computational cost.Meanwhile,it can mitigate the imbalance issue between training data and real-world face images due to the progressive refinement mechanism,and thus boost the generalization ability of the model.Specifically,in the training phase,our model relies solely on a small amount of labeled data.The earlier prediction serves as guidance for following refinement.In addition,our shared-parameter cascaded block employs a recurrent mechanism,allowing it to be applied multiple times for optimization without increasing network parameters.Quantitative and qualitative evaluations on benchmark datasets are conducted to show that our Cas-FNE can faithfully maintain facial details and reveal its superiority over state-of-the-artmethods.The code is available at https://github.com/AutoHDR/CasFNE.git.
文摘The incremental radiosity method has been shown to be an efficienttechnique for providing global illumination in dynamic environments as it exploitstemporal coherence in object space. This paper presents an accelerated incremental radiosity algorithm, which is based on a dynamically followed partial matrix.This not only reduces the computation cost in determining incremental form-factorswhen the geometrical relationships between objects are constantly changing, but alsosimplifies the management of user interaction with comparatively little storage cost.
