Investigación

Artículos en revistas científicas

Publicación (Tesis doctorales)

The digitization of pathology departments in hospitals around the world is now a reality. The current commercial solutions applied to digitize histopathological samples consist of a robotic microscope with an RGB-type camera attached to it. This technolog

Saha, Anup

Resumen:

Nearly everybody in the modern era frequently uses video compression technology in the form of watching TV, under-demand content, video streaming, multimedia content on social networks, self-made videos, etc. Video compression technology is essential to transmit videos through the network and store them in an efficient way. The very first of video compression technology was introduced and patented by Ray Davis Kell in 1929. It consisted on store differences of frames as the variations in the inductance of transformer coils. However, researchers from industry and academia started working on standardising video compression technology from the 19th century. But, it was necessary to wait till 1984 for the standardization of the first video coding standard, H.120, proposed by the International Telegraph and Telephone Consultative Committee (CCITT). CCITT further developed the H.261 standard by cooperating with the International Telecommunication Union (ITU) in 1988. Afterwards, all the video coding standards followed the same fundamental video coding recommendations with a hybrid coding inner structure, where transform coding is combined with intra-prediction and, later on, inter-prediction. Since then, and almost on average every six years, a new standard was released due to the increased demand for higher quality videos and with the aim of achieving significant bitrate savings for comparable subjective video quality. During the last decades, the demand for higher quality videos (full and ultra high definition) has increased rapidly due to the wide availability of consumer electronics devices. To meet this demand, the Joint Video Experts Team (JVET) released the Versatile Video Coding (VVC) standard in July 2020. VVC provides up to 50% bitrate savings compared to the previous High Efficiency Video Coding (HEVC) standard by keeping the same video quality. However, this reduction in bitrate causes a major increase in computational complexity, especially for real-time video processing on embedded devices with limited resources. To overcome this added computational complexity challenge for obtaining real-time performance, it is highly required to exploit the available parallelism either in software and hardware. In this scenario, this doctoral research focused on the development of a hybrid approach to accelerate the newer VVC-based decoders over heterogeneous platforms using finegrain parallelism, coarse-grain parallelism, and hardware accelerators. Here, data-level parallelism technique Single Instruction Multiple Data (SIMD), Central Processing Unit (CPU) + Graphics Processing Unit (GPU) based implementation were integrated with native coding default features and coarse-grain parallelism. First, a detailed analysis was performed by profiling the state-of-the-art VVC decoder to identify the most computationally demanding blocks. Thus, coarse-grain and fine-grain profile of the decoder were carried out. Moreover, parallel processing abilities of different VVC decoder blocks were analysed to apply optimisation techniques. Secondly, different platforms and open source video decoders were chosen and migrated to the selected platforms. Then, SIMD optimisation was efficiently implemented on the latest version of the decoders on heterogeneous platforms. Therefore, a hybrid approach was implemented on heterogeneous platforms, both SIMD optimisation and heterogeneous CPU+GPU based implementation were used in parallel with native coding default features and coarse-grain parallelism. These processes were repeated for different versions of the decoders on different platforms. This work was verified by conducting a wide set of experiments with fifteen sequences with different configurations included in the common test conditions standard set. Lastly, a design methodology was synthesised using the experience of previous optimization. This methodology proposes some recommendations to accelerate the optimization process of video decoders with different platforms. It has been used to optimize other decoder implementation (OpenVVC) with good results and reducing the development effort.

Áreas de investigación: