In this paper, the concept of non-desynchronizing bits (NDBs) is defined in the context of H.264 video as a bit whose inversion does not cause desynchronization at the bitstream level or change the number of decoded macroblocks. We established that, on the whole, NDBs make up about a third (about 30%) of a bitstream, and that their flipping effect on visual quality is mostly insignificant. In most cases (90%), the PSNR value obtained when modifying an NDB is very close to the intact value. The performance of the proposed non-desync-based decoding framework, which retains a corrupted packet, under the condition of not causing desynchronization, has been compared to the JM-FC and a state-of-the-art concealment approach using the STBMA approach, and on average, respectively, provides 3.5 dB and 1.42 dB gain over them. Index Terms-video transmission, H.264, syntax elements, nondesynchronizing bit, concealment This work was funded by the Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant.
3D and multi-view 360-degree videos offer 3D immersive visual experience of the real world on head-mounted displays (HMDs). They are used in extended reality applications (including virtual reality, augmented reality and mixed reality) such as gaming, entertainment, education and medicine. However, these applications require a very high rate of compression and a great deal of computational resources. Compression efficiency and coding complexity are highly influenced by 360 projections used to map spherical frames to rectangular frames. To study the effect of this mapping process, in this paper, we evaluate the performance of various projections in 3D and multi-view 360-degree videos in terms of coding efficiency, quality and complexity. The evaluation is conducted for two coding scenarios: stereo coding scenario where the encoder considers the interview dependency for disparity estimation and simulcast coding scenario where two views are encoded separately. This performance evaluation approach provides valuable insights on using 360-degree projections in 3D and multi-view coding.
The latest video coding standards, H.264 and H.265, are highly vulnerable in error-prone networks. Reconstructed packets may exhibit significant degradation in terms of PSNR and visual quality. This paper presents a novel list decoding approach exploiting the receiver side user datagram protocol (UDP) checksum. The proposed method identifies the possible locations of errors by analyzing the pattern of the calculated UDP checksum. This permits to considerably reduce the number of candidate bitstreams in comparison to conventional list decoding approaches. When a packet composed of N bits contains a single-bit error, instead of considering N candidate bitstreams, as is the case in conventional list decoding approaches, the proposed approach considers N/32 candidate bitstreams, leading to a reduction of 97% of the number of candidates. For a two-bit error, the reduction increases to 99.6%. The method's performance is evaluated using H.264 and H.265 test model software. Our simulation results reveal that, on average, the error was corrected perfectly 80 to 90% of the time (the original bitstream was recovered). In addition, the proposed approach provides, on average, a 2.79 dB gain over frame copy (FC) error concealment using the Joint Model (JM) and a 3.57 dB gain over our implementation of FC error concealment in the HEVC Test Model (HM).
In this paper, we introduce a novel cyclic redundancy check (CRC)-based single error correction method which we apply to robust H.264 Baseline video decoding. Unlike stateof-the-art methods, the proposed correction algorithm does not require lookup tables as it determines the error location based on binary operations using the computed link layer CRC syndrome. Since multiple errors can lead to the same CRC syndrome as a single error, verification of the corrected packet is performed through a non-desynchronizing bits validation (NDBV), which forwards only compliant packets to the video decoder. Simulations on the H.264 Baseline profile show an average gain of 3.04 dB and 2.36 dB over stateof-the-art spatio-temporal error concealment (STBMA) and NDBV+STBMA reconstruction methods, respectively, at a residual bit error rate of 10 −6 .
This paper presents a novel list decoding approach exploiting the receiver side user datagram protocol (UDP) checksum. The proposed method identifies the possible locations of errors in the packet by analyzing the calculated UDP checksum value at the receiver side. This makes it possible to considerably reduce the number of candidate bitstreams in comparison to conventional list decoding approaches. When a packet composed of N bits contains a single bit in error, instead of considering N candidate bitstreams, as is the case in conventional list decoding approaches, the proposed approach considers N/32 candidate bitstreams, leading to a 97% reduction in the number of candidates. Our simulation results on H.264 compressed sequences reveal that, on average, the error is corrected perfectly 80% of the time, and thus, the original bitstream is fully recovered when the first valid candidate is considered as the best candidate. In addition, the proposed approach provides, on average, a 2.78 dB gain over the error concealment approach used by the H.264 reference software, as well as 1.31 dB and 1.51 dB gains over the state-of-the-art error concealment and HO-MLD approaches, respectively.
Abstract-In this work, we study the performance of the maximum likelihood decoding (MLD) approach in error-resilient video sequences to establish the performance improvement of this method compared to well-known error concealment approaches. In particular, we consider various interactions between error resilience coding and error concealment/correction. The error resilience methods under consideration include random intra macroblock updating and weighted error resilience. For error concealment/correction, we consider (i) the frame copy (FC), (ii) spatio-temporal boundary matching error concealments (STBMA), and (iii) MLD error correction. Our experimental results show that the best performance is achieved when the MLD interacts with weighted error resilience. Together, they yield, on average, about a 2 dB gain over using FC error concealment with weighted error resilience and a 1 dB gain over STBMA with identical error resilience. Furthermore, MLD with error resilience can be more than 10 dB better than FC without error resilience in certain cases.
The recent popularity of using deep learning models for the forecasting of time series calls for methods to not only predict the target but also measure the uncertainty of the prediction accurately. Working with time series requires reliable and stable forecasters. An essential component of the reliability of machine learning (ML) and deep learning (DL) models is the estimation of the uncertainty. In this work, we address building and characterizing time series forecasters, including N-Beats, Long Short-Term Memory (LSTM) and Multilayer Perceptron (MLP) against the Naive model, and define the confidence margins, and uncertainty for the selected model.All the implementations are conducted in Python programming language. Random sampling is performed to avoid overfitting. Our target field data is North American Service Provider data sets (NASP). Among the implemented models, the MLP model is selected to measure the uncertainty and confidence level, and the Monte Carlo dropout, which approximates Bayesian uncertainty, is applied during inference to render the implementation of uncertainty calculations. Quantile Regression is also implemented on the MLP algorithm as a baseline to predict the confidence intervals and to evaluate our strategy for estimating uncertainty. To establish reliable uncertainty estimation in time series predictions, we performed uncertainty calibration. Motivated by recent developments in Expected Uncertainty Calibration Error (UCE), we modified the uncertainty calculated by the probabilistic Bayesian estimations. Detailed experiments and architectures of the solution are presented.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.