Speeding up motion estimation algorithms on CUDA technology

Cheng, Ronghui; Yang, Eryan; Liu, Ting

doi:10.1109/primeasia.2010.5604952

Cited by 10 publications

(6 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, most of the parallel ME research work is on many-core concentrates on the full-search method, which is inherently highly parallel. In Chen and Hang [15]; Cheng et al [16]; Lee and Oh [17]; Monteiro et al [18], the parallel fullsearch method is implemented on the GPU platform, and about 10-100x speed-ups are, respectively, obtained compared with the serial full-search method on single core of a CPU. Although the speed-up of the full-search method is high on GPU platform, its performance advantage is not obvious compared with serial fast search method in HEVC or H.264/AVC on single core of a CPU.…”

Section: Related Workmentioning

confidence: 99%

A Highly Parallel and Scalable Motion Estimation Algorithm with GPU for HEVC

Xue

Ren

et al. 2017

Scientific Programming

View full text Add to dashboard Cite

We propose a highly parallel and scalable motion estimation algorithm, named multilevel resolution motion estimation (MLRME for short), by combining the advantages of local full search and downsampling. By subsampling a video frame, a large amount of computation is saved. While using the local full-search method, it can exploit massive parallelism and make full use of the powerful modern many-core accelerators, such as GPU and Intel Xeon Phi. We implanted the proposed MLRME into HM12.0, and the experimental results showed that the encoding quality of the MLRME method is close to that of the fast motion estimation in HEVC, which declines by less than 1.5%. We also implemented the MLRME with CUDA, which obtained 30-60x speed-up compared to the serial algorithm on single CPU. Specifically, the parallel implementation of MLRME on a GTX 460 GPU can meet the real-time coding requirement with about 25 fps for the 2560 × 1600 video format, while, for 832 × 480, the performance is more than 100 fps.

show abstract

Section: Related Workmentioning

confidence: 99%

A Highly Parallel and Scalable Motion Estimation Algorithm with GPU for HEVC

Xue

Ren

et al. 2017

Scientific Programming

View full text Add to dashboard Cite

show abstract

“…Yang et al optimized the exhaustive search with shared memory [31]. Cheng et al evaluated exhaustive search, diamond search, and four-step search in the GPU [32]. Attempts to use the GPU for deblocking and intra coding were also reported in [33] and [34].…”

Section: Encoding Using Gpumentioning

confidence: 99%

HEVC Encoding Optimization Using Multicore CPUs and GPUs

et al. 2015

IEEE Trans. Circuits Syst. Video Technol.

View full text Add to dashboard Cite

Although the High Efficiency Video Coding (HEVC) standard significantly improves coding efficiency of video compression, it is unacceptable even in offline applications to spend several hours compressing 10 seconds of High Definition (HD) video. In this paper, we propose using a multi-core Central Processing Unit (CPU) and an off-the-shelf Graphics Processing Unit (GPU) with 3,072 streaming processors (SPs) for HEVC fast encoding so that the speed optimization does not result in loss of coding efficiency. There are two key technical contributions in this paper. First, we propose an algorithm that is both parallel and fast for the GPU, which can utilize 3,072 SPs in parallel to estimate the motion vector of every Prediction Unit (PU) in every combination of the Coding Unit (CU) and PU partitions. Furthermore, the proposed GPU algorithm can avoid coding efficiency loss caused by the lack of a motion vector predictor (MVP). Second, we propose a fast algorithm for the CPU, which can fully utilize the results from the GPU to significantly reduce the number of possible CU and PU partitions without any coding efficiency loss. Our experimental results show that compared with the reference software, we can encode high resolution video that consumes 1.9% of the CPU time and 1.0% of the GPU time, with only a 1.4% rate increase.

show abstract

“…MB level parallelism is exploited where the MBs in the frame are evenly partitioned among the available processing cores. Note that the parallel ES algorithms proposed in [37][38][39][40][41], and implemented using OpenCL [37,38] and CUDA [39][40][41], are also based on MB level parallelism along with search point parallel processing to compute the cost of each search point. Simulation results are given interms of the average number of fitness function evaluations per lab for a given frame based on the first 100 frames of every sequence.…”

Section: Comparison With Existing Parallel Me Algorithmsmentioning

confidence: 99%

“…As a result, the proposed algorithm provides tremendous speedup if implemented on modern high performance computing (HPC) platforms ranging from multicore/many-core machine architectures to graphics processing units to supercomputers. In the literature, there have been several attempts to parallelize motion estimation [37][38][39][40][41][42]. Several works have proposed applying GPUs for motion estimation.…”

Section: Introductionmentioning

confidence: 99%

“…Implementation of the ES motion estimation algorithm with OpenCL has been proposed in [37] and [38]. In [39], implementations of the ES algorithm, the diamond search (DS) algorithm, and the four-step search (4SS) algorithms in CUDA have been proposed. A parallel implementation of the ES algorithm on the GPU using CUDA is also proposed in [40,41] along with a parallel solution for multi-core processors using the Open Message Passing (OpenMP) library and a distributed solution for cluster/grid machines using the Message Passing Interface (MPI) library [41].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation