Title: Complexity prediction of hardware and software video transcoding in the cloud

Authors: Taieb Chachou, Sid Ahmed Fezza, Wassim Hamidouche, Ghalem Belalem, Hadi Amirpour

[PDF]

Abstract: Today, video content constitutes a significant portion of internet traffic. This video can be viewed by a wide range of devices with varying characteristics and under different network conditions. Video transcoding is a crucial mechanism for adapting video content to this diverse array of devices and bandwidth requirements while ensuring the best possible user experience. However, video transcoding is a computationally intensive process, requiring scalable infrastructure like cloud computing to efficiently handle the complexity and volume of tasks. In this paper, we propose a novel method to predict transcoding time across different types of platforms (CPU and GPU) and codecs (H.264/AVC, H.265/HEVC).

Unlike existing approaches that focus mainly on CPU-based transcoding, the proposed model explicitly considers hardware-accelerated (GPU) transcoding, where accelerators significantly influence video transcoding performance in cloud computing.

The predicted transcoding time can be utilized to optimize the scheduling of transcoding tasks in cloud computing, helping to ensure optimal load balancing and minimize total transcoding time while maintaining the highest video quality. The proposed solution consists of two essential phases: (i) dataset construction and (ii) model construction.  The first phase involves video selection, segmentation, and video transcoding. The second phase focuses on analyzing the most important features that influence the prediction of transcoding time and developing a machine learning-based model for accurate video transcoding time prediction. Experimental results demonstrate that the XGBoost model achieves superior prediction accuracy across both software and hardware codecs, achieving a global coefficient of determination of R²~=~0.993 when evaluated on the complete dataset, which includes video segments transcoded using H.264/AVC and H.265/HEVC codecs on CPU and GPU platforms. This performance represents an improvement of approximately 7.45% compared to state-of-the-art methods.

Title: Quality-Complexity Trade-off for Sustainable Media Delivery

Authors: Hadi Amirpour, Christian Herglotz, Lingfeng Qu, Wei Zhou, Christian Timmerer

Event: QoMEX 2026, Cardiff, UK, June 29th – July 3rd, 2026

Abstract: Sustainable media delivery increasingly requires joint optimization across perceptual quality, bitrate, and computational cost, yet codec comparisons are often reported only in rate-distortion terms without accounting for energy and (en/de)coding time overheads at scale. This paper analyzes quality–rate–computational cost trade-offs using a large-scale dataset. We first quantify the dominant drivers of bitrate, VMAF, and (en/de)coding user time via interpretable regression models, showing that codec and resolution explain a substantial fraction of the observed variance.  We then characterize local sensitivities of bitrate and (en/de)coding user time to incremental increases in VMAF using interpolation in the quality domain and finite-difference derivatives, providing a content-agnostic view of how much additional bitrate and compute, and consequently energy expenditure, is required per unit of quality improvement. To evaluate practical savings, we compute Bjøntegaard Delta metrics relative to a libx264 reference, revealing that large BD-Rate gains can coincide with substantial penalties in (en/de)coding user time, particularly for most recent video coding standards such as Versatile Video Coding (VVC). Finally, we formulate multi-objective configuration selection as a Binary Linear Program (BLP) that selects one operating point per video by trading perceptual quality against bitrate and (en/de)coding user time; across different weight regimes, the selected codec-resolution-frame-rate distributions shift coherently with system priorities.

Title: EPS: Efficient Patch Sampling for Video Overfitting in Deep Super-Resolution Model Training

Authors: Yiying Wei, Hadi Amirpour, Jong Hwan Ko, and Christian Timmerer

Abstract: Leveraging the overfitting property of deep neural networks (DNNs) is trending in video delivery systems to enhance video quality within bandwidth limits. Existing approaches transmit overfitted super-resolution (SR) model streams for low-resolution (LR) bitstreams, which are used to reconstruct high-resolution (HR) videos at the decoder. Although these approaches show promising results, the huge computational costs of training a large number of video frames limit their practical applications. To overcome this challenge, we propose an efficient patch sampling method named EPS for video SR network overfitting, which identifies the most valuable training patches from video frames.

To this end, we first present two low-complexity Discrete Cosine Transform (DCT)-based spatial-temporal features to measure the complexity score of each patch directly. By analyzing the histogram distribution of these features, we then categorize all possible patches into different clusters and select training patches from the cluster with the highest spatial-temporal information. The number of sampled patches is adaptive based on the video content, addressing the trade-off between training complexity and efficiency.

Our method reduces the number of training patches by 75.00\% to 91.69\%, depending on the resolution and number of clusters, while preserving high video quality and greatly improving training efficiency. Our method speeds up patch sampling by up to 82.1$\times$ compared to the state-of-the-art patch sampling technique (EMT).