Reliable co-speech motion generation requires precise motion representation and consistent structural priors across all joints. Existing generative methods typically operate on local joint rotations, which are defined hierarchically based on the skeleton structure. This leads to cumulative errors during generation, manifesting as unstable and implausible motions at end-effectors. In this work, we propose GlobalDiff, a diffusion-based framework that operates directly in the space of global joint rotations for the first time, fundamentally decoupling each joint’s prediction from upstream dependencies and alleviating hierarchical error accumulation. To compensate for the absence of structural priors in global rotation space, we introduce a multi-level constraint scheme. Specifically, a joint structure constraint introduces virtual anchor points around each joint to better capture fine-grained orientation. A skeleton structure constraint enforces angular consistency across bones to maintain structural integrity. A temporal structure constraint utilizes a multi-scale variational encoder to align the generated motion with ground-truth temporal patterns. These constraints jointly regularize the global diffusion process and reinforce structural awareness. Extensive evaluations on standard co-speech benchmarks show that GlobalDiff generates smooth and accurate motions, improving the performance by 46.0% compared to the current SOTA under multiple speaker identities.

Co-speech gesture generation must carefully integrate common rhythmic motion with rare yet essential semantic gestures. In this work, we propose SemTalk for holistic co-speech gesture generation with frame-level semantic emphasis. Our key insight is to separately learn base motions and sparse motions, and then adaptively fuse them. In particular, coarse2fine cross-attention module and rhythmic consistency learning are explored to establish rhythm-related base motion, ensuring a coherent foundation that synchronizes gestures with the speech rhythm. Subsequently, semantic emphasis learning is designed to generate semantic-aware sparse motion, focusing on frame-level semantic cues. Finally, to integrate sparse motion into the base motion and generate semantic-emphasized co-speech gestures, we further leverage a learned semantic score for adaptive synthesis. Qualitative and quantitative comparisons on two public datasets demonstrate that our method outperforms the state-of-the-art, delivering high-quality co-speech motion with enhanced semantic richness over a stable base motion.

Limited by the nature of the low-dimensional representational capacity of 3DMM, most of the 3DMM-based face reconstruction (FR) methods fail to recover high-frequency facial details, such as wrinkles, dimples, etc. Some attempt to solve the problem by introducing detail maps or nonlinear operations, however, the results are still not vivid. To this end, we in this paper present a novel hierarchical representation network (HRN) to achieve accurate and detailed face reconstruction from a single image. Specifically, we implement the geometry disentanglement and introduce the hierarchical representation to fulfill detailed face modeling. Meanwhile, 3D priors of facial details are incorporated to enhance the accuracy and authenticity of the reconstruction results. We also propose a deretouching module to achieve better decoupling of the geometry and appearance. It is noteworthy that our framework can be extended to a multi-view fashion by considering detail consistency of different views. Extensive experiments on two single-view and two multi-view FR benchmarks demonstrate that our method outperforms the existing methods in both reconstruction accuracy and visual effects. Finally, we introduce a high-quality 3D face dataset FaceHD-100 to boost the research of high-fidelity face reconstruction. The project homepage is at https://younglbw.github.io/HRN-homepage/.

Body reshaping is an important procedure in portrait photo retouching. Due to the complicated structure and multifarious appearance of human bodies, existing methods either fall back on the 3D domain via body morphable model or resort to keypoint-based image deformation, leading to inefficiency and unsatisfied visual quality. In this paper, we address these limitations by formulating an end-to-end flow generation architecture under the guidance of body structural priors, including skeletons and Part Affinity Fields, and achieve unprecedentedly controllable performance under arbitrary poses and garments. A compositional attention mechanism is introduced for capturing both visual perceptual correlations and structural associations of the human body to reinforce the manipulation consistency among related parts. For a comprehensive evaluation, we construct the first large-scale body reshaping dataset, namely BR-5K, which contains 5,000 portrait photos as well as professionally retouched targets. Extensive experiments demonstrate that our approach significantly outperforms existing state-of-the-art methods in terms of visual performance, controllability, and efficiency. The dataset is available at our website: https://github.com/JianqiangRen/FlowBasedBodyReshaping.

To obtain sufficient observation data and simulate higher-precision crop yields, a crop yield simulation scheme was built based on the WOrld FOod STudies (WOFOST) crop growth model and a 4-D ensemble square root filter (4-DEnSRF) assimilation algorithm, and the time series of the leaf area index (LAI) retrieved by optical and synthetic aperture radar (SAR) networking data was introduced into the crop yield estimation scheme. Taking Shenzhou County, Hebei Province, as the study area, using the field-measured data as verification data, the regional application of winter wheat yield estimation was effectively carried out with a grid size of 500 m. Comparisons were made between the simulated yields based on different networked data of three key phenologies of winter wheat. The regional yield estimation results revealed an$R^{2}$and normalized root mean squared error (NRMSE) between the simulated yield based on optical LAIs and the field-measured yield of 0.517 and 17.60%, respectively, while the$R^{2}$and NRMSE between the simulated yield based on networked optical-SAR LAIs filtered by the Gaussian filtering algorithm (GFA) and the field-measured yield were 0.573 and 12.98%, respectively. From the comparisons between the simulated yields based on networked data of different combinations of key phenologies, the$R^{2}$and NRMSE between the simulated yield based on the introduced SAR LAI at the jointing stage and the field-measured yield were 0.437 and 21.49%, respectively, and were higher correlation among the three modes of networked data of different combinations of key phenologies. The winter wheat yield simulation results showed that the introduction of SAR LAIs at key crop growth stages (especially the jointing and booting stage) as outer observation data had a mild impact on the value of simulated winter wheat yield. Moreover, Gaussian filtering could reduce errors caused by multisource networked data to a certain extent. Thus, it can be concluded that using some radar images instead of optical images to retrieve LAI and assimilating multisource remotely sensed LAI into the crop model to simulate crop yield could enhance the reliability and robustness of the crop yield simulation system to some extent.

Neural style transfer has drawn considerable attention from both academic and industrial field. Although visual effect and efficiency have been significantly improved, existing methods are unable to coordinate spatial distribution of visual attention between the content image and stylized image, or render diverse level of detail via different brush strokes. In this paper, we tackle these limitations by developing an attention-aware multi-stroke style transfer model. We first propose to assemble self-attention mechanism into a style-agnostic reconstruction autoencoder framework, from which the attention map of a content image can be derived. By performing multi-scale style swap on content features and style features, we produce multiple feature maps reflecting different stroke patterns. A flexible fusion strategy is further presented to incorporate the salient characteristics from the attention map, which allows integrating multiple stroke patterns into different spatial regions of the output image harmoniously. We demonstrate the effectiveness of our method, as well as generate comparable stylized images with multiple stroke patterns against the state-of-the-art methods.

To further improve the accuracy of remote sensing classification and land-cover recognition at sub-pixel level, a novel sub-pixel mapping (SPM) model was first proposed by introducing the concept of pixel aggregation degree (PAD) which could simulate the spatial distribution of small-sized land-cover. In the proposed novel SPM model, based on the distribution of sub-pixel random initialization, PAD algorithm was optimized sub-pixel distribution to obtain final SPM results. Using a Sentinel-2 remote sensing data, related SPM experiments were performed to verify both accuracy and effect of PAD SPM model. The experimental results indicated that the SPM accuracy based on PAD model were superior to the classification results of the K-mean and the SPM results of traditional spatial attraction model. It was shown that the PAD model had certain feasibility and applicability which provided a new idea to better break the limitations of remote sensing image spatial resolution, and was beneficial to the subsequent research and application of remote sensing image.

Online automatic measurement of traffic state parameters has important significance for intelligent transportation surveillance. The video-based monitoring technology is widely studied today but the existing methods are not satisfactory at processing speed or accuracy, especially for traffic scenes with traffic congestion or complex road environments. Based on technologies of computer vision and pattern recognition, this paper proposes a novel measurement method that can detect multiple parameters of traffic flow and identify vehicle types from video sequence rapidly and accurately by combining feature points detection with foreground temporal-spatial image (FTSI) analysis. In this method, two virtual detection lines (VDLs) are first set in frame images. During working, vehicular feature points are extracted via the upstream-VDL and grouped in unit of vehicle based on their movement differences. Then, FTSI is accumulated from video frames via the downstream-VDL, and adhesive blobs of occlusion vehicles in FTSI are separated effectively based on feature point groups and projection histogram of blob pixels. At regular intervals, traffic parameters are calculated via statistical analysis of blobs and vehicles are classified via a K-nearest neighbor (KNN) classifier based on geometrical characteristics of their blobs. For vehicle classification, the distorted blobs of temporary stopped vehicles are corrected accurately based on the vehicular instantaneous speed on the downstream-VDL. Experiments show that the proposed method is efficient and practicable.

To obtain spatial feature distributions from the mixed pixels of remote sensing images and increase the accuracy of land-cover classification and recognition, a double-calculated spatial attraction model (DSAM) based on the combination of spatial attraction model (SAM) and pixel swapping model (PSM) is presented and verified by introducing the law of universal gravitation to describe the attraction between pixels. In DSAM, SAM was used to improve the initialization algorithm of PSM, and the optimization algorithm of PSM was improved accordingly. Using a SPOT-5 remote sensing image, related subpixel mapping (SPM) experiments were performed to verify the SPM effect of DSAM and to test its accuracy. The experimental results indicated that the DSAM SPM results were superior to the SPM results of SAM and PSM. DSAM was proven effective and applicable for the SPM of remotely sensed images, and the model can improve the accuracy of SPM and landcover classification.

With the sustaining economic development in China, the timely, accurate and objective agricultural production information service has been highly demanded by the central and provincial governments. China Agricultural Remote Sensing Monitoring System (CHARMS) is an operational agricultural monitoring system in the Mnistry of Agriculture of China to meet this demand. Wheat, corn and rice, are the three main grain crops in the world, and they are also most import grain crops in China. According to the statistic data, the acreage and production of these crops is more than 85% of that of all grain crops in China. After accurate monitoring of these crops, we will know the overall grain production in China. Other important crops, including soybean, cotton, canola and sugarcane are also among the monitoring target crops in the system. Besides, the grassland productivity, degradation of grassland and grass-livestock balance are also monitored and evaluated in the system. The system of CHARMS consists of following components: the database sub-system, crop acreage change monitoring module, crop yield estimation module, crop growth monitoring module, soil moisture monitoring module, disaster monitoring module and information service module. In CHARMS, remote sensing, CIS, GNSS and conventional methods have been integrated and sampling methods have been employed to monitor crop production in China, especially in the key regions. With the progressively monitoring of growth and drought in crop planting regions, crop-specific yield prediction models have been set up to estimate the yield and production in the nation. The monitoring results have been submitted to Mnistry of Agriculture and other agriculture production authorities, to provide decision-support information for their everyday management and policy-making.

In recent years, combining spatial and timely remote sensing data and crop growth model is an important way to improve accuracy of crop growth simulation and crop growth monitoring. In this paper, global optimization algorithm SCE-UA (Shuffled Complex Evolution method University of Arizona) was used to integrate remotely sensed leaf area index (LAI) with EPIC crop growth model to simulate regional winter wheat yield and other field management information such as sowing date, plant density and net nitrogen fertilizer application rate in Huanghuaihai Plain in China. Final results showed that average relative error of estimated winter wheat yield was 1.81% and RMSE was 0.208 t/ha. Compared with the actual observation data, average relative error of simulated plant density and net nitrogen fertilization application rate was -7.95% and -8.88% respectively and absolute error of simulated sowing date was only 1 day. These above accuracy of simulated results could meet requirements of crop monitoring at regional scale. It was proved that integrating remotely sensed LAI with EPIC model based on SCE-UA for simulation of crop growth condition and crop yield was feasible.

Spatial and temporal mismatches between coarse resolution output of global climate models (GCMs) and fine resolution data requirements of crop models are the major obstacles for assessing the site-specific climatic impacts of climate change on the production of winter wheat. Based on the output of IPCC AR4 model and observation data, statistical downscaling of precipitation, minimum temperature, and maximum temperature in North China was analyzed. With the combination crop model and climate mode, the effects of climate change on the winter wheat production of North China were simulated. Some conclusions from the study might be drawn as follows: Under the IPCC-B1 Scenario, the length of winter wheat growing season in North China would be shortened from 2010 to 2099, and its yield would be decreased.

Solar irradiation is a way to characterize the climate of particular region, and used in tourism and agriculture. The HELIOSAT method of deriving solar irradiation from FY- 2C geostationary satellite images was evaluated for north China using situ data. The results clear show that the FY-2C data can be used for mapping surface solar irradiation over north China. The FY-2C data is useful for where the measured solar irradiation is not available.

In order to acquire the spatial information of winter wheat harvest index (HI), depending on the crop growth profile of time-series MODIS-NDVI to calculate mean slope of NDVI curve and accumulated NDVI at stage of before anthesis and after anthesis, the authors made full use of remote sensing information and structured two parameters HINDVI-k and HINDVI-SUM according to the definition of HI of crop. Then relationships between the two parameters and field measured HI of winter wheat were established respectively. After validation of retrieved winter wheat HI, it was shown that the accuracy of the retrieved HI of winter wheat was high and satisfied at large scale in study region of Huanghuaihai Plain in China. The mean relative errors of the retrieved HI were 3.60% and 2.40% and RMSE were 0.04 and 0.02 respectively. It was proved that the method of structuring parameters of HINDVI-k and HINDVI_SUM and extracting harvest index for winter wheat based on timeseries MODIS-NDVI was reasonable and feasible.

Assimilating external data into crop growth model to improve accuracy of crop growth monitoring and yield estimation has been being a research hotspot in recent years. In this paper, the global optimization algorithm SCE-UA (Shuffled Complex Evolution method-University of Arizona) was used to integrate remotely sensed leaf area index (LAI) with crop growth model EPIC to simulate regional yield, sowing date, plant density and net nitrogen fertilizer application rate of summer maize in Huanghuaihai Plain. The final results showed that average relative error of estimated summer maize yield was 4.37% and RMSE was 0.44t/ha. Meanwhile, compared with actual observation and investigation data, average relative error of simulated sowing date, plant density and net N fertilization application rate was 1.85%, -7.78% and -10.60% respectively. These above accuracy of simulated results could meet the need of crop monitoring at regional scale. It was proved that integrating remotely sensed LAI with EPIC model based on global optimization algorithm SCE-UA for simulation of crop growth condition and crop yield was feasible.

In this paper, the authors had a research on the winter wheat yield estimation using retrieved LAI from remote sensing in typical 11 counties in Huanghuaihai Plain in North China. In order to improve the quality of data and reduce error of yield estimation, Savitzky-Golay filter (S-G filter) was used to smooth the NDVI series to reduce influences of cloud contamination and abnormal data. At the same time, Gaussian model was used to simulate daily crop LAI to get average LAI at each growth stage. Using these average LAI, the authors established relationships between LAI and yield of winter wheat at main growth stage. After optimization of yield estimation model, the best period of time and best model was selected out. Finally, the authors depended on retrieved LAI from MODIS-NDVI to estimate winter wheat yield. The results showed that average relative error was 1.21% and that RMSE was 257.33 kg ha−1comparing predicted yield with ground truth data. We draw a conclusion that we could accurately get winter wheat yield about 20–30 days ahead of harvest time.

Assimilating external data into a crop growth model to improve accuracy of crop growth monitoring and yield estimation has been a research focus in recent years. In this paper, the shuffled complex evolution (SCE-UA) global optimization algorithm was used to assimilate field measured LAI into EPIC model to simulate yield, sowing date and nitrogen fertilizer application amount of summer maize in Huanghuaihai Plain in China. The results showed that RMSE between simulated yield and field measured yield of summer maize was 0.84 t ha-1and the R2was only 0.033 without external data assimilation. While the performances of EPIC model of simulating yield, sowing date and nitrogen fertilizer application amount of summer maize was better through assimilating field measured LAI into the EPIC model. The RMSE of between simulated yield and field measured yield of summer maize was 0.60 t ha-1and the R2was 0.5301. The relative error between simulated sowing date and real sowing date of summer maize was 2.28%. On the simulation of nitrogen fertilizer application rate, the relative error was -6.00% compared with local statistical data. These above accuracy could meet the need of crop growth monitoring and yield estimation at regional scale. It proved that assimilating field measured LAI into crop growth model based on SCE-UA optimization algorithm to monitor crop growth and estimate crop yield was feasible.

Crop yield data is a key indicator for national food security and sustainable development of society. Winter wheat is one of the most important main crops and Huanghuaihai Plain is the most important productive region in North China. So, the authors had a research on regional yield prediction for winter wheat based on crop biomass estimation using multi-source data in the Plain. In this paper, the quantitative relationship between crop biomass and yield was mainly used to estimate crop yield. The total crop biomass is cumulative biomass in the critical crop growth stages which are significant to final yield. Crop biomass was calculated by net primary production (NPP) model. Then NPP was converted into crop biomass considering factors such as carbon content of winter wheat, harvest index (HI), water content of grain, etc. When NPP was calculated, the model: NPP =ε * fPAR * PAR was used. The photosynthetically active radiation (PAR, 400–700nm) was calculated using ultraviolet (UV) reflectance data at 370 nm which came from the Total Ozone Mapping Spectrometer (TOMS). The fraction of photosynthetically active radiation (fPAR) could be retrieved from MODIS NDVI imagery using linear regression between fPAR and NDVI. The biological conversion efficiency of dry matter (ε) was calculated depending on relationship between ε and its affecting factors such as average temperature, rainfall and relative moisture of soil. Finally, the method and established relationships were applied in a larger region. Through validation, the absolute mean error of estimated yield was 232.7 kg.ha−1and relative error of estimated yield was 4.28%. Judged from the results of validation, the method based on crop biomass estimation using multi-source data was effective and practical in yield estimation of winter wheat in a larger region.