Search

Built on the explosive advancement of cloud and telecommunication technologies, Gaming-as-a-Service (GaaS) or cloud gaming system is expected to revolutionize the traditional multi-billion video game market in the near future. This wave is analogous to the rise of live-video-streaming-based-Netflix to replace conventional DVD rental business for movies and TVs. In practice, a successful GaaS platform need to operate in a transparent mode without requiring substantial efforts from both content providers and end users, and offer the pristine quality of experience (QoE) at an affordable cost. Our analysis suggests that GaaS provisioning cost can be reduced significantly by enforcing the game video rendering and streaming at a lower resolution (so as to increase the user concurrency in the cloud and reduce the streaming bandwidth over the network). However, streaming video at a lower resolution may deteriorate the QoE. To maintain the client QoE at the level using the default-native resolution for streaming or even enhance it, we introduce the learned resolution scaling (LRS), which leverages the computational capabilities at clients/edges to restore/improve the reconstructed image/video quality via stacked deep neural networks (DNN). We integrate this LRS into a commercialized GaaS platform - AnyGame, to study its efficiency and complexity quantitatively. Extensive real-life experiments have shown that LRS-powered AnyGame offers the state-of-the-art performance, and the lower operational cost, paving the road for a potential success of GaaS over the Internet. Additionally, we dive into proposed LRS via ablation studies to further demonstrate its consistent performance, including the discussions on trade-off between efficiency and complexity, alternative training sets, etc.

Viewport-Dependent Streaming (VDS) is a preferred way in practice to deliver the omnidirectional videos, of which a High-Quality (HQ) scale is applied for the content in current viewport but a Low-Quality (LQ) scale elsewhere. Quality adaptation or refinement happens after users stabilize their fixations to a new viewport. In this article, we formulate this as a perceptually optimized quality adaptation problem to maximize the Quality of Experience (QoE) for the refinement from a LQ scale to another HQ level within a specific duration under the given network constraint. With our developed perceptual model considering the adaptation quality for VDS of omnidirectional videos, we first provide baseline solutions numerically, demonstrating the noticeable subjective improvements of model-driven solution against the heuristic selection based approach. We also propose a novel viewport prediction algorithm based on the Hidden Markov Model (HMM), and experimental results show that it significantly outperforms the relevant methods with better prediction accuracy. We then improve the adaptation strategy with proposed viewport prediction-based data prefetching, leading to better visual perception than the baseline system at the same bandwidth constraint. Generally, prefetching the content of predicted next viewport one second ahead of its playback time, would lead to more than 8% Bjontegaard Delta Rate (BD-Rate) gain.

Due to the subjective annotation and the inherent interclass similarity of facial expressions, one of key challenges in Facial Expression Recognition (FER) is the annotation ambiguity. In this paper, we proposes a solution, named DMUE, to address the problem of annotation ambiguity from two perspectives: the latent Distribution Mining and the pairwise Uncertainty Estimation. For the former, an auxiliary multi-branch learning framework is introduced to better mine and describe the latent distribution in the label space. For the latter, the pairwise relationship of semantic feature between instances are fully exploited to estimate the ambiguity extent in the instance space. The proposed method is independent to the backbone architectures, and brings no extra burden for inference. The experiments are conducted on the popular real-world benchmarks and the synthetic noisy datasets. Either way, the proposed DMUE stably achieves leading performance., Accepted by CVPR21

Background: Surgical ligation and endovascular embolization have been recommended for type II congenital extrahepatic portosystemic shunt (CEPS); however, no consensus has been reached. This study was designed to compare the safety and efficacy of surgical ligation and endovascular embolization for the treatment of type II CEPS. Methods: In this retrospective study, 23 consecutive patients diagnosed with type II CEPS between March 2011 and April 2019 were divided into either a surgical group (n=13; 41.5±19.9years) or the interventional group (n =10; 44.9±19.7years). The surgical group underwent laparoscopic surgical ligation of the shunt alone or ligation of the shunt and splenic artery and/or vein. The interventional group underwent endovascular embolization using microcoils, detachable coils and vascular plug. Results: All 23 patients received a one-step shunt closure, and their clinical symptoms were significantly improved within 3 months post-procedure and without recurrence during follow-up. The serum ammonia levels in both groups decreased after the procedure and dropped to normal level at 6 to 12 months post-procedure. Compared with baseline, the portal vein diameter in interventional group increased significantly at 3-, 6-, 12-, and 36-month post-occlusion (P=0.01 for all). The procedure time was shorter in interventional group (127.0±43.2minutes) than surgical group (219.8±56.7minutes; P Conclusion: Both surgical ligation and endovascular embolization are effective in the treatment of type II CEPS. Endovascular embolization has the advantages of shorter procedure time, and less intraoperative blood loss. The ligation of the portosystemic shunt and splenic artery and vein is feasible with apparent safety, and it could avoid a second surgical treatment.

Hyperspectral images (HSIs) contain the response of each pixel in different spectral bands, which can be used to effectively distinguish various objects in complex scenes. While HSI cameras have become low cost, algorithms based on it have not been well exploited. In this paper, we focus on a novel topic, weakly-supervised semantic segmentation in cityscape via HSIs. It is based on the idea that high-resolution HSIs in city scenes contain rich spectral information, which can be easily associated to semantics without manual labeling. Therefore, it enables low cost, highly reliable semantic segmentation in complex scenes. Specifically, in this paper, we theoretically analyze the HSIs and introduce a weakly-supervised HSI semantic segmentation framework, which utilizes spectral information to improve the coarse labels to a finer degree. The experimental results show that our method can obtain highly competitive labels and even have higher edge fineness than artificial fine labels in some classes. At the same time, the results also show that the refined labels can effectively improve the performance of existing semantic segmentation algorithms. The combination of HSIs and semantic segmentation proves that HSIs have great potential in high-level visual tasks for automatic driving.

In this paper, we present a large-scale detailed 3D face dataset, FaceScape, and propose a novel algorithm that is able to predict elaborate riggable 3D face models from a single image input. FaceScape dataset provides 18,760 textured 3D faces, captured from 938 subjects and each with 20 specific expressions. The 3D models contain the pore-level facial geometry that is also processed to be topologically uniformed. These fine 3D facial models can be represented as a 3D morphable model for rough shapes and displacement maps for detailed geometry. Taking advantage of the large-scale and high-accuracy dataset, a novel algorithm is further proposed to learn the expression-specific dynamic details using a deep neural network. The learned relationship serves as the foundation of our 3D face prediction system from a single image input. Different than the previous methods, our predicted 3D models are riggable with highly detailed geometry under different expressions. The unprecedented dataset and code will be released to public for research purpose., Comment: Accepted to CVPR 2020

Increased drought frequency in Australia is a pressing concern for scholars. In 2018, a severe drought in eastern Australia was recorded by the Emergency Events Database (EM-DAT). To investigate the main causes and impacts of this drought across southeastern Australia, this work presents an overview of the drought mechanism and depicts its evolutionary process. The Standardized Precipitation Evapotranspiration Index (SPEI) from the Global Drought Monitor was used to identify the drought event and characterize its spatiotemporal distribution. The Normalized Difference Vegetation Index (NDVI) and the sun-induced chlorophyll fluorescence (SIF) were used to investigate the drought impacts on vegetation growth. In addition, the effects of drought response measures on Sustainable Development Goals (SDGs) were analyzed. Our results showed that the exceptional drought occurred across southeastern Australia from April to December, and it was most severe in July, owing to an extreme lack of precipitation and increase in temperature. Moreover, we identified profound and long-lasting impacts of the drought on NDVI and SIF levels, especially for cropland. Furthermore, we also found that SIF was superior to NDVI in detecting drought impacts. This study advised on how to formulate timely and effective drought-response measures and supports sustainable socioeconomic development in Australia.

With the explosive increase of image data, the efficiency of both image compression and retrieval becomes unprecedentedly significant. However, these two tasks are usually isolated executed, which waste great computational resources in large-scale image applications. In this work, we propose a joint framework called CodedRetrieval, which can find a general feature expression for both compression and retrieval based on neural network. Additionally, a two stage training strategy is designed to achieve better balance between the two distinct tasks. Experimental results show that our method can achieve competitive perfomance on both compression and retrieval comparing to classic methods, while saving great amount of computation time.

We propose a MultiScale AutoEncoder(MSAE) based extreme image compression framework to offer visually pleasing reconstruction at a very low bitrate. Our method leverages the "priors" at different resolution scale to improve the compression efficiency, and also employs the generative adversarial network(GAN) with multiscale discriminators to perform the end-to-end trainable rate-distortion optimization. We compare the perceptual quality of our reconstructions with traditional compression algorithms using High-Efficiency Video Coding(HEVC) based Intra Profile and JPEG2000 on the public Cityscapes and ADE20K datasets, demonstrating the significant subjective quality improvement., Comment: Accepted to IEEE VCIP 2019 as an oral presentation

Cloud radio access networks (C-RANs) have been proposed to provide high energy-spectral efficiency and high data rate at low cost. In this correspondence paper, we introduce a novel C-RAN deployment scheme that is complementary to traditional cellular radio networks. The key idea is to install the remote radio heads (RRHs) of the C-RAN in a load-balanced way to lighten the transmission burden of the backhauls between the RRHs and the centralized baseband unit pool, and to activate the RRHs on demand so as to reduce system-wide energy consumption. Numerical results show that our proposed method performs well in practical scenarios. It provides QoS-guaranteed performance with relatively low capital expenditure and operating expenditure.

This article proposes a transparent gaming (TG) cloud system that allows users to play any popular high-end desktop game on the fly over the Internet. Toward this goal, we have introduced the TG-SHARE technology to share the underlying hardware capabilities, particularly for the GPU and the dedicated compression acceleration unit (XCODER). TG-SHARE utilizes offthe- shelf consumer GPUs without resorting to expensive proprietary GPU virtualization technology (e.g., GRID from NVIDIA). XCODER adapts the compression based on the network dynamics, learned gaming behaviors, and hardware resources to significantly reduce bandwidth consumption. Google's webRTC protocol is integrated to offer real-time interaction and ubiquitous access from heterogeneous devices. Compared to the existing cloud gaming vendor using the GRID technology, our TG-SHARE not only reduces the expense per user (i.e., 75 percent hardware cost reduction, 20-40 percent network cost reduction), but also improves the quality of experience with higher rate of frames per second (i.e., 2 x FPS).

3D motion estimation via feature point matching is an important issue in machine vision. However, most existing methods are not good enough due to various defects, such as being sensitive to matching error, and hard to give an appropriate threshold for feature point selection. To solve these problems, a mathematical model is obtained from massive experiments to describe how the quantity and quality of matched feature points influence the motion estimation accuracy. This model can be used to select the most appropriate feature points for motion parameter calculation, hence threshold setting is avoided while estimation performance is optimized. Experimental results show that the proposed algorithm can improve the accuracy of 3D motion estimation by up to 50% with little effect on computation time. Furthermore, it is especially suitable for real time applications as there is no need to set threshold manually.

We have developed a Looking-Ahead system to facilitate the future video frame prediction via deep learning, which is of practical value in the domain like autonomous driving etc. The overall problem is decomposed into cascaded optical flow prediction and subsequent predictive frame post-processing for quality refinement. A pyramid flow calculation across existing frames is used to efficiently infer the motion of target frame; while a universal inpainting network is applied to restore those motion-induced occluded pixels. Compared with those published methods, our Looking-Ahead offers the state-of-the-art performance measured objectively with better Peak-Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM), and more appealing reconstructions.

Networked video applications, e.g., video conferencing, often suffer from poor visual quality due to unexpected network fluctuation and limited bandwidth. In this paper, we have developed a Quality Enhancement Network (QENet) to reduce the video compression artifacts, leveraging the spatial and temporal priors generated by respective multi-scale convolutions spatially and warped temporal predictions in a recurrent fashion temporally. We have integrated this QENet as a standard-alone post-processing subsystem to the High Efficiency Video Coding (HEVC) compliant decoder. Experimental results show that our QENet demonstrates the state-of-the-art performance against default in-loop filters in HEVC and other deep learning based methods with noticeable objective gains in Peak-Signal-to-Noise Ratio (PSNR) and subjective gains visually.

Network dynamics, such as bandwidth fluctuation and unexpected latency, hurt users’ quality of experience (QoE) greatly for media services over the Internet. In this work, we propose a neural adaptive transport (NAT) framework to tackle the network dynamics for Internet-scale interactive media services. The entire NAT system has three major components: a learning based cloud overlay routing (COR) scheme for the best delivery path to bypass the network bottlenecks while offering the minimal end-to-end latency simultaneously; a residual neural network based collaborative video processing (CVP) system to trade the computational capability at client-end for QoE improvement via learned resolution scaling; and a deep reinforcement learning (DRL) based adaptive real-time streaming (ARS) strategy to select the appropriate video bitrate for maximal QoE. We have demonstrated that COR could improve the user satisfaction from 5% to 43%, CVP could reduce the bandwidth consumption more than 30% at the same quality, and DRL-based ARS can maintain the smooth streaming with < 50% QoE improvement, respectively.

We develop a model that expresses the joint impact of spatial resolution $s$ and JPEG compression quality factor $q^{f}$ on immersive image quality. The model is expressed as the product of optimized exponential functions of these factors. The model is tested on a subjective database of immersive image contents rendered on a head mounted display. High Pearson correlation and Spearman correlation (>0.95) and small relative root mean squared error (

Robust visual object tracking is a challenging task in computer vision. Recently correlation filter-based trackers (CFTs) have aroused increasing interests because of the good performance and high efficiency. However, most feature representations for CFTs are not discriminative enough, which makes the trackers unreliable in complicated and changing scenarios. To address the problem, this paper presents an adaptive multi-feature fusion method based on kernelized correlation filter (KCF) framework. First we select HOG, LBP and grayscale feature for fusion to obtain more complementary and powerful feature. Then we propose a novel multi-feature fusion strategy, and adaptively calculate the feature’s fusion weight using probability separability criterion. The experimental results show that our method not only achieves better accuracy compared with existing features for KCF tracker, but also achieves state-of-the-art performance when running at 87 frames per second.

Immersive video offers the freedom to navigate inside the virtualized environment. Instead of streaming the entire bulky content, a viewport or field of view (FoV) adaptive streaming is preferred. We often stream the high-quality content within current viewport, but degraded-quality representation elsewhere, so as to reduce the network bandwidth consumption. We then could refine the quality when focusing to a new FoV. Therefore, in this work, we have attempted to model the perceptual response of the quality variations (through adapting the quantization and spatial resolution) with respect to the refinement duration, and reach at a product of two closed-form exponential functions that well explain the joint quantization and resolution induced quality impact. Analytical model is also cross-validated using another set of data with both Pearson and Spearman's rank rank correlations over 0.98. Our work would be devised to guide the bandwidth-quality optimized immersive video streaming.

Inspired by recent advances in deep learning, we present the DeepCoder — a Convolutional Neural Network (CNN) based video compression framework. We apply separate CNN nets for predictive and residual signals respectively. Scalar quantization and Huffman coding are employed to encode the quantized feature maps (fMaps) into binary stream. We use the fixed 32 × 32 block in this work to demonstrate our ideas, and performance comparison is conducted with the well-known H.264/AVC video coding standard with comparable rate-distortion performance. Here distortion is measured using Structural Similarity (SSIM) because it is more close to perceptual response.

We present an efficient multi-image super resolution (MISR) method. Our solution consists of a L1-norm optimized reconstruction scheme for super resolution (SR), and a three-layer convolutional network for artifacts removal, in a concatenated fashion. Such a two-stage method achieves excellent performance, which outperforms the existing state-of-the-art SR methods in both subjective and objective measurements (e.g., 5 to 7 dB improvements on popular image database using PSNR metric).

Floods are among the most expensive natural hazards experienced in many places of the world and can result in heavy losses of life and economic damages. The objective of this study is to analyze flood inundation in ungauged basins by performing near-real-time detection with flood extent and depth based on multi-source remote sensing data. Via spatial distribution analysis of flood extent and depth in a time series, the inundation condition and the characteristics of flood disaster can be reflected. The results show that the multi-source remote sensing data can make up the lack of hydrological data in ungauged basins, which is helpful to reconstruct hydrological sequence; the combination of MODIS (moderate-resolution imaging spectroradiometer) surface reflectance productions and the DFO (Dartmouth Flood Observatory) flood database can achieve the macro-dynamic monitoring of the flood inundation in ungauged basins, and then the differential technique of high-resolution optical and microwave images before and after floods can be used to calculate flood extent to reflect spatial changes of inundation; the monitoring algorithm for the flood depth combining RS and GIS is simple and easy and can quickly calculate the depth with a known flood extent that is obtained from remote sensing images in ungauged basins. Relevant results can provide effective help for the disaster relief work performed by government departments.

Drought is a complex natural phenomenon that can cause reduced water supplies and can consequently have substantial effects on agriculture and socioeconomic activities. The objective of this study was to gain a better understanding of the spatial-temporal variation characteristics of vegetative drought and its relationship with meteorological factors in China. The Vegetation Condition Index (VCI) dataset calculated from NOAA/AVHRR images from 1982 to 2010 was used to analyse the spatial-temporal variation characteristics of vegetative drought in China. This study also examined the trends in meteorological factors and their influences on drought using monitoring data collected from 686 national ground meteorological stations. The results showed that the VCI appeared to slowly rise in China from 1982 to 2010. From 1982 to 1999, the VCI rose slowly. Then, around 2000, the VCI exhibited a severe fluctuation before it entered into a relatively stable stage. Drought frequencies in China were higher, showing a spatial distribution feature of “higher in the north and lower in the south”. Based on the different levels of drought, the frequencies of mild and moderate drought in four geographical areas were higher, and the frequency of severe drought was higher only in ecologically vulnerable areas, such as the Tarim Basin and the Qaidam Basin. Drought was mainly influenced by meteorological factors, which differed regionally. In the northern region, the main influential factor was sunshine duration, while the other factors showed minimal effects. In the southern region and Tibetan Plateau, the main influential factors were sunshine duration and temperature. In the northwestern region, the main influential factors were wind velocity and station atmospheric pressure.

Drought is a type of natural disaster that has the most significant impacts on agriculture. Regional drought monitoring based on remote sensing has become popular due to the development of remote sensing technology. In this study, vegetation condition index (VCI) data recorded from 1982 to 2010 in agricultural areas of China were obtained from advanced very high resolution radiometer (AVHRR) data, and the temporal and spatial variations in each drought were analyzed. The relationships between drought and climate factors were also analyzed. The results showed that from 1982 to 2010, the agricultural areas that experienced frequent and severe droughts were mainly concentrated in the northwestern areas and Huang-Huai Plain. Moreover, the VCI increased in the majority of agricultural areas, indicating that the drought frequency decreased over time, and the decreasing trend in the southern region was more notable than that in the northern region. A correlation analysis showed that temperature and wind velocity were the main factors that influenced drought in the agricultural areas of China. From a regional perspective, excluding precipitation, the climate factors had various effects on drought in different regions. However, the correlation between the VCI and precipitation was low, possibly due to the widespread use of artificial irrigation technology, which reduces the reliance of agricultural areas on precipitation.

Video adaptation has been considered as a promising technique to tackle challenging problems in pervasive media applications. However, the style of video representation and description in current framework is too simple to accommodate diversified usages, especially for realtime application environment. In this paper, an intermediate video description termed intermedia is proposed, which is generated by off-line pre-processing the original video signal and includes abundant information of video content. On the basis of intermedia, a novel framework for video adaptation can be implemented to handle multiple video applications. The framework has low complexity operations not only in signal level, but also in structural level, and in joint level. Texture, motion and structural characteristic are included in intermedia, so that the framework can tackle a number of different adaptation applications, e.g., bit-rate conversion, temporal resolution reduction, video summarization browsing, and multi-dimensional adaptation. Experimental results demonstrate that intermedia can support multiple video adaptations rapidly while maintaining coding efficiency.

In order to improve the recognition of license plate texts in the real traffic surveillance video, super-resolution reconstruction (SR) method is applied to reconstruct a high-resolution (HR) image from consecutive frames in the video sequence. Current SR methods almost focus on small local translation and rotation, which limits the reconstruction of fast moving or significantly zooming objects, such as license plate images in surveillance video. In this paper, we combine Fourier-Mellin transform (FMT) and Vandewalle's algorithm into a new technique to improve the registration accuracy of license plate image. FMT is introduced for scaling estimation, while Vandewalle's algorithm is utilized for rotation and translation estimation. Additionally, the image reconstruction is carried out by the projection onto convex sets (POCS) method. The experiments on simulated and real image sequences are carried out respectively, and the results demonstrate that our approach can achieve better performance on reconstructing a HR license plate image.

Sparse unmixing is a promising approach in a semisupervised fashion by assuming that the observed signatures of a hyperspectral image can be expressed in the form of linear combination of only a few spectral signatures (endmembers) in an available spectral library. Simultaneous orthogonal matching pursuit (SOMP) algorithm is a typical simultaneous greedy algorithm for sparse unmixing, which involves finding the optimal subset of signatures for the observed data from a spectral library. But the numbers of endmembers selected by SOMP are still larger than the actual number, and the nonexisting endmembers will have a negative effect on the estimation of the abundances corresponding to the actual endmembers. This paper presents a variant of SOMP, termed backtracking-based SOMP (BSOMP), for sparse unmixing of hyperspectral data. As an extension of SOMP, BSOMP incorporates a backtracking technique to detect the previous chosen endmembers’ reliability and then deletes the unreliable endmembers. Through this modification, BSOMP can select the true endmembers more accurately than SOMP. Experimental results on both simulated and real data demonstrate the effectiveness of the proposed algorithm.

Objective: To investigate the significance of intrafascial neurovascular bundle spare technique in radical cystectomy. Methods: Between March 2010 and December 2011, a total of 26 bladder cancer patients were treated with radical cystectomy, and intrafascial neurovascular bundle spare technique was applied in all these patients. Mean age of 26 patients was 56.1 y (45 - 66). Among 26 patients, 21 cases were in stage T2, 5 cases were in stage T3. All patients choose Orthotopic neobladder as urinary diversion manner. We use intrafascial neurovascular bundle spare technique, dissect between prostatic fascial and prostatic capsule, spare neurovascular bundle. Operating time, blood loss, complications, continence and sexual function 3 months after surgery were recorded. Results: In all patients, mean operating time was 328 min, mean blood loss was 316 ml. Only 4 patients need transfusion during surgery. 1 case of urinary fistula was found after surgery, and spontaneously cured 10 days after surgery. 1 case of bowel obstruction was found, and was cured by conservative therapy. 4 cases of incontinence were found 3 months after surgery. 18 patients had a nomal erectile function 3 months after surgery. Conclusions: Intrafascial neurovascular bundle spare technique can safely and effectively reserve neurovascular bundle in radical cystectomy. Patients can reserve continence and erectile function by this technique.

Video adaptation has been considered as a promising technique to tackle challenging problems in pervasive multimedia applications. However, the styles of video representation and description in existing framework are not flexible enough to adapt diversified application environment. In this paper, we propose a novel solution based on intermediate description, which can support fast multiple video adaptation operations in signal level (e.g., temporal resolution reduction, bit-rate adaptation), structural level (e.g., random access of any shots, fast preview of any key-frames or thumbnails), as well as joint level. Experimental results show that the presented solution can support the operations in real-time environment, while maintaining coding efficiency, which demonstrates its feasibility and effectiveness.

The aim of this study was to determine whether treatment of patients with immune thrombocytopenic purpura (ITP) with intravenous immunoglobulin (IVIg) is associated with a modification in the antiplatelet glycoprotein (GP) antibodies (Abs). Fourteen patients with ITP (11 females and 3 males, mean age 36.6 years, range 18–72) received one to four IVIg treatment courses. The preparation used was ISIVEN that was given in a dose of 2 g/kg body weight in a 5-day schedule and in monthly intervals. Levels of IgG, IgM and IgA isotypes of Abs to GPs IIb/IIIa and Ib/IX were measured before the treatment, and before and after each treatment course. Two patients did not respond to IVIg, 6 had a temporary response, 5 had a sustained response and 1 patient responded well to the treatment but was lost to follow-up. The patients had a high prevalence of serum Abs directed against GPs IIb/IIIa and Ib/IX before the treatment, and the mean IgG isotype levels of both Abs increased after each treatment course, and decreased again before the following course began. Whenever high Ab levels of either isotype (>10 U/ml) were detected before the treatment, they were significantly decreased before the last treatment course. The elevated levels of IgG Abs to IIb/IIIa and Ib/IX after every course are probably a result of displacement of these Abs from Fc receptors by the IVIg, rather than of exogenous infusion of these Abs contained within the IVIg, whereas the decrease in high Ab levels after a few treatment courses results from the immunomodulatory effects of IVIg: suppression of Ab formation, and the presence of anti-idiotypes.

The infectivity of a node is determined by its actual betweenness bk in the epidemic model based on traffic-flow other than degree k which is different from the classical epidemic models. Utilizing the mean-field theory, we propose a modified SIS epidemic model based on traffic-flow. With this model, taking MIP route as an example, we re-study the relationship among spreading probability β, traffic generation rate λ, epidemic threshold βc, the stationary density of infected nodes ρ. Both theoretical analysis and experimental results show that βc is the ratio of the mean of the actual betweenness bk > to its mean square bk2 >, when network topology and route strategy are given. Moreover, the stationary density of infected notes ρ behaves as power-law exponent with the reciprocal of the product of the spreading probability β , the traffic generation rate λ and the mean of the actual betweenness bλ=1 >.