Given an arbitrary face image and speech clip, the proposed work attempts to generate talking video with accurate lip synchronization. Existing works either do not consider temporal dependency across frames thus yielding abrupt facial movement or are limited generation of for a specific person lacking generalization capacity. We propose novel conditional recurrent network that incorporates both audio features in unit dependency. To achieve image- video-realism, pair spatial-temporal...

Object detection models shipped with camera-equipped edge devices cannot cover the objects of interest for every user. Therefore, incremental learning capability is a critical feature robust and personalized object system that many applications would rely on. In this paper, we present an efficient yet practical system, RILOD, to incrementally train existing model such it can detect new classes without losing its old classes. The key component RILOD novel algorithm trains end-to-end one-stage...

This study comprehensively investigated the impact of indoor carbon dioxide (CO2 ) concentration on sleep quality. Three experimental conditions (800, 1900, 3000 ppm) were created in chambers decorated as bedroom and other environmental parameters that may influence quality under strict control. Sleep 12 subjects (6 men 6 women) was monitored for 54 consecutive days through questionnaire physiological measures. Both subjective results showed decreased significantly with increase CO2...

Traditional β-eucryptite (LiAlSiO4) is renowned for its unique characteristics of low thermal expansion and high temperature stability, making it an ideal material precision instruments aerospace applications. In this study, was fabricated into aerogel structure through the sol-gel process supercritical drying method, using alumina sol as a cost-effective precursor. The synthesized has such negative (-7.85×10-6 K-1), density (0.36 g/cm3) large specific surface area (70.9 m2/g). X-ray...

The existing methods of learning to rank often ignore the relationship between ranking features. If them can be fully utilized, performance improved. Aiming at this problem, an approach that combines a multi-head self-attention mechanism with Conditional Generative Adversarial Nets (CGAN) is proposed in paper, named *GAN-LTR. improves some design ideas Information Retrieval Networks (IRGAN) framework applied web search, and new network model constructed by integrating convolution layer,...

The human eye cannot perceive small pixel changes in images or videos until a certain threshold of distortion. In the context video compression, Just Noticeable Difference (JND) is smallest distortion level from which can difference between reference and distorted/compressed one. Satisfied-User-Ratio (SUR) curve complementary cumulative distribution function individual JNDs viewer group. However, most previous works predict each point SUR by using features both source compressed with...

In HTTP adaptive live streaming applications, video segments are encoded at a fixed set of bitrate-resolution pairs known as bitrate ladder. Live encoders use the fastest available encoding configuration, referred to preset, ensure minimum possible latency in encoding. However, an optimized preset and number CPU threads for each instance may result (i) increased quality (ii) efficient utilization while For low encoders, speed is expected be more than or equal framerate. To this light, paper...

This study improves a shear-based rutting model for asphalt concrete (AC) layers and calibrates the with field data. With dynamic modulus-based material parameters, laboratory prediction was improved determined by wheel tracking test full-scale accelerated pavement test. Through survey on several in-service pavements, calibrated to be applied AC layers. In model, ratio of maximum shear stress strength introduced combine design structural design. The speed correction coefficient new...

Given an arbitrary face image and speech clip, the proposed work attempts to generating talking video with accurate lip synchronization while maintaining smooth transition of both facial movement over entire clip. Existing works either do not consider temporal dependency on images across different frames thus easily yielding noticeable/abrupt or are only limited generation for a specific person lacking generalization capacity. We propose novel conditional network where audio input is treated...

Portrait segmentation has gained more and attractions in recent years due to the popularity of selfie images. Compared general semantic problems, portrait focuses on facial areas with higher requirements especially over boundaries. To improve performance segmentation, we propose a boundary-sensitive deep neural network (BSN) for better accuracy among BSN introduces three novel techniques. First, an individual mask is proposed by dilating contour line assigning boundary pixels multi-class...

In HTTP Adaptive Streaming (HAS), a video is encoded at multiple bitrate-resolution pairs, referred to as representations, which enables users choose the most suitable representation based on their network connection. To optimize set of pairs and improve Quality Experience (QoE) for users, it utmost importance measure quality representations. VMAF highly reliable metric used in HAS assess However, practice, using optimization can be very time-consuming process, infeasible live streaming...

Observer screening and subject opinion score recovery is essential for collecting a reliable QoE database. This paper proposes new method, ZREC*, which uses Z-scores to estimate bias, inconsistency, content ambiguity. Additionally, we propose Mean Opinion Score (MOS) Percentile (POS) scheme based on the three estimated parameters. ZREC does not fully reject subjects, rather adjust their coefficients in MOS/POS recovery, allowing more efficient use of data collection. The parameters are...

Just Noticeable Difference (JND) model developed based on Human Vision System (HVS) through subjective studies is valuable for many multimedia use cases. In the streaming industries, it commonly applied to reach a good balance between compression efficiency and perceptual quality when selecting video encoding recipes. Nevertheless, recent state-of-the-art deep learning JND prediction relies largescale ground truth that expensive time consuming collect. Most of existing datasets contain...

This paper addresses the effect of gradation on high temperature performance and water stability asphalt rubber mixture that contains Evotherm. The mixtures was selected for analysis with warm-mix additive. volumetric properties hot-mix were investigated by Marshall Test, rutting test freeze-thaw split test. results indicated aggregate ratio played a critical role in determination feasibility process. AR-AC-13 not suitable process due to over binder content. effective improve AR-SMA at lower...

Just Noticeable Difference (JND) and Satisfied User Ratio (SUR) has been widely investigated for compressed image video to use the least resources (e.g., storage bandwidth) without damaging Quality of Experience (QoE) end users. However, current JND subjective test methodologies are extremely time consuming due large range encoding parameters. Besides, state-of-the-arts SUR/JND prediction models get non-negligible error limited masking effect features. To this end, we first proposed a...

In video streaming applications, a fixed set of bitrate-resolution pairs (known as bitrate ladder) is typically used during the entire session. However, an optimized ladder per scene may result in (i) decreased storage or delivery costs or/and (ii) increased Quality Experience. This paper introduces Just Noticeable Difference (JND)-aware perscene prediction scheme (JASLA) for adaptive video-on-demand applications. JASLA predicts jointly resolutions and corresponding constant rate factors...