Panos K. Papadopoulos

Slice-based parallelization in HEVC encoding: Realizing the potential through efficient load balancing

The new video coding standard HEVC (High Efficiency Video Coding) offers the desired compression performance in the era of HDTV and UHDTV, as it achieves nearly 50% bit rate saving compared to H.264/AVC. To leverage the involved computational overhead, HEVC offers three parallelization potentials namely: wavefront parallelization, tile-based and slice-based. In this paper we study slice-based parallelization of HEVC using OpenMP on the encoding part. In particular we delve on the problem of proper slice sizing to reduce load imbalances among threads. Capitalizing on existing ideas for H.264/AVC we develop a fast dynamic approach to decide on load distribution and compare it against an alternative in the HEVC literature. Through experiments with commonly used video sequences, we highlight the merits and drawbacks of the tested heuristics. We then improve upon them for the case of Low-Delay by exploiting GOP structure. The resulting algorithm is shown to clearly outperform its counterparts achieving less than 10% load imbalance in many cases.

On planning the adoption of new video standards in social media networks: a general framework and its application to HEVC

In recent years, we have witnessed an explosion in the growth of social media networks, powered by the proliferation of handheld smart devices with high processing capabilities and a plethora of sensors including high-resolution cameras. A key component of information exchange in such networks, accounting for the majority of network traffic, is video. Currently, the de facto video coding standard in use is H.264/AVC which was sufficient in addressing the challenges posed by HD more than a decade ago, but is less than efficient in the new era of 4K smart device cameras and 8K TV screens. Given that newer standards exist and are capable of achieving higher compression rates at the same quality compared to H.264/AVC, we envision that within the next few years, the related industry will shift toward one of the newer video coding standards. For a social media network, such a transition poses manifold challenges, one of them being the need to transcode previous content in the newly adopted standard. In this paper, we illustrate a framework for performing such transition in a smooth manner. The framework, algorithms and strategies developed are applicable, perhaps with minor changes, regardless of the targeted standard for adoption. We detail on framework components through simulation experiments, using as a yardstick the adoption of high efficiency video coding. Results demonstrate that depending on the targeted social platform, different strategies should be applied, while the cost and benefits of the paradigm shift may vary significantly.

Heuristics for tile parallelism in HEVC

HEVC has emerged as the new video coding standard promising increased compression ratios compared to its predecessors. This performance improvement comes at a high computational cost. For this reason, HEVC offers three coarse grained parallelization potentials namely, wave front, slices and tiles. In this paper we focus on tile parallelism which is a relatively new concept with its effects not yet fully explored. Particularly, we investigate the problem of partitioning a frame into tiles so that in a resulting one on one tile-CPU core assignment the cores are load balanced, thus, maximum speedup can be achieved. We propose various heuristics for the problem with a focus on low delay coding and evaluate them against state of the art approaches. Results demonstrate that particular heuristic combinations clearly outperform their counterparts in the literature.

Performance Evaluation of Batch Encodings in HEVC Using Slice Level Parallelism

HEVC has emerged as the new video standard to replace H.264/AVC. Although the new standard is able to achieve significantly higher compression ratios compared to the older one, it entails high computational demands. To alleviate the problem, parallelism opportunities are offered by the standard at different levels namely, slice, tile and wave front. In this paper we evaluate the performance of slice level parallelism in HEVC encoding. Although significant previous work exists on slice parallelization, particularly for H.264/AVC, usually the experimental evaluation was limited to characterizing the performance when encoding one sequence at a time. Our focus is to evaluate slice parallelism when a batch of encoding jobs is submitted to the system, with CPU core requirements potentially exceeding the available resources.

Adaptive Tile Parallelization for Fast Video Encoding in HEVC

As large multimedia providers rely more and more on Cloud resources to perform video coding and transcoding, designing fast and efficient coders that take advantage of parallelization, particularly for the new video standard HEVC, is of paramount importance. Tiles were introduced in HEVC to provide an alternative parallelization granularity compared to the previous standard H.264/AVC. The premise was that with the rectangular tile shapes, spatial correlation between samples could be better exploited compared to slices, leading to increased coding efficiency. While a significant amount of research was done in parallelizing video coding, few works exist on tile parallelization in HEVC. In this paper we tackle the problem of balancing the CPU core load by dynamically adapting tile sizes. It is shown that the proposed adaptive method leads to significant reduction of load imbalances and consequently, achieves a better speedup compared to static, uniform CTU-tile assignment. Furthermore, these gains come at no cost quality wise.

Combining tile parallelism with slice partitioning in video coding

Tiles and slices provide different frame partitioning options. While they can both be used for video coding parallelization, tiles offer better scalability to the number of available processors, especially as far as video quality is concerned, e.g., in the HEVC case. On the other hand, slices can be useful in video transmission. Since slices can be defined as a series of consecutive (raster order) tiles, properly balancing them can lead to viable trade-offs between parallelization and transmission requirements. In this paper we study the combined problem of tile and slice partitioning with the goals of maximizing the achievable parallelism speedup, while minimizing size difference among slices. These goals might conflict with each other, while producing multiple Pareto frontier solutions can introduce additional time overhead. For these reasons, we map the two-function optimization problem to a single one, using constant weighting and develop algorithms that perform tile resizing and slice definition so as to optimize the composite target function. Experiments with common class A and class B test sequences and the reference HEVC encoder (HM), reveal that compared to static uniform tile partitioning and to literature alternatives that resize tiles in order to increase parallelization speedup, the proposed algorithm achieves considerable gains in slice balancing, while also improving speedup over the static approach. Furthermore, these performance merits come with negligible overhead to the encoding process.

On Improving the Speedup of Slice and Tile Level Parallelism in HEVC Using AVX2

HEVC has emerged as the new video coding standard promising improved compression ratios (for the same quality) by up to 50% compared to H.264/AVC. To achieve this performance HEVC requires increased computational overhead compared to its predecessor. For this reason parallelism is used, usually at a coarse grained level, e.g., per slice or tile. In this paper we turn our attention towards further speeding up the HEVC encoding process by combining coarse grained parallelism with fine grained, in the form of AVX2 instructions implementing SIMD parallelism at SAD (Sum of Absolute Difference) and SSE (Sum of Squared Error) calculations. Experimental evaluation with common test video sequences illustrates that an additional reduction (in encoding time) of roughly 11% on average, compared to standalone coarse grained parallelism is achievable, leading in many cases to superlinear speedup.

Mobivoke: A Mobile System Architecture to Support off School Collaborative Learning Process.

The collaborative learning paradigm offers one of the most solid approaches to increase the participation, interest and knowledge level of pupils (typical achieving and/or learning disabled students) during the educational process. Recent advances in the field have offered a plethora of tools to facilitate collaboration during school time. Nevertheless, the possibility of applying the collaborative learning principles together with personalized exercise/project assignments (whenever deemed necessary) during off school hours is often overlooked. Motivated by the fact that most pupils/students nowadays have access to smart mobile devices, e.g., tablets, in this paper a system architecture (Mobivoke) is proposed that enables the coupling of individual devices into a social group and offers the means to build applications for orchestrating and monitoring the off school learning process in a collaborative manner.

RAC: a remote application calling framework for coordination of mobile apps

Mobile applications (apps) have become part of our everyday life with a constantly increasing market. Of particular interest are apps aiding planning and collaboration among family members, or between co-workers. The architecture of such apps usually involves some Cloud storage medium, through which group members post and retrieve data. Naturally, all participants must have the same app installed in their devices for collaboration to be possible. In this paper we investigate an alternative option instead of app collaboration which is based on remote application calling (RAC). Under the RAC framework, a trusted source is able to invoke application actions at other peoples devices, without necessarily owning himself the application it handles. We discuss RAC design and implementation related issues, focusing on Android devices. The usability of our approach is demonstrated through two widely used apps: alarm clock and map.