multimedia AI | BestHub

Bilibili Tech

Nov 7, 2023 · Artificial Intelligence

Technical Overview of Bilibili Vision Toolkit (BVT): Architecture, Features, and FFmpeg Filter Integration

The Bilibili Vision Toolkit (BVT) is a C++ SDK that unifies multimedia AI algorithms through a low‑coupling core, modular dynamic libraries, and a multi‑engine backend, enabling configurable DAG pipelines, asynchronous parallel execution, and seamless FFmpeg filter integration for high‑performance, cross‑platform video processing.

BVTC++FFmpeg

0 likes · 15 min read

Technical Overview of Bilibili Vision Toolkit (BVT): Architecture, Features, and FFmpeg Filter Integration

Tencent Advertising Technology

Dec 20, 2022 · Artificial Intelligence

Modeling Advertising Attractiveness: Data Analysis, Pairwise Learning, and DeepFM Optimization

This article presents a comprehensive study on estimating video ad attractiveness by analyzing 3‑second completion rates, proposing pairwise MLP and DeepFM models, introducing hierarchical sampling and multimodal features, and demonstrating practical deployment improvements in material recommendation and ad ranking.

advertisingattractivenessdeepFM

0 likes · 16 min read

Modeling Advertising Attractiveness: Data Analysis, Pairwise Learning, and DeepFM Optimization

Tencent Advertising Technology

Nov 22, 2022 · Artificial Intelligence

Tencent Advertising Multimedia AI: Systems, Technologies, and Business Applications

The article presents Tencent Advertising's comprehensive multimedia AI platform—including content understanding, intelligent creation, automated review, fingerprinting, and a large multimodal model—detailing their architectures, key techniques, and the significant performance and efficiency gains achieved across the ad ecosystem.

Ad Generationadvertising AIautomated review

0 likes · 18 min read

Tencent Advertising Multimedia AI: Systems, Technologies, and Business Applications

Alimama Tech

Nov 2, 2022 · Artificial Intelligence

Optimizing GPU Utilization for Multimedia AI Services with high_service

The article presents high_service, a high‑performance inference framework that boosts GPU utilization in multimedia AI services by separating CPU‑heavy preprocessing from GPU inference, employing priority‑based auto‑scaling, multi‑tenant sharing, and TensorRT‑accelerated models to eliminate GIL bottlenecks, reduce waste, and adapt to fluctuating traffic, with future work targeting automated bottleneck detection and further CPU‑GPU offloading.

GPU utilizationHigh Performance ComputingTensorRT

0 likes · 19 min read

Optimizing GPU Utilization for Multimedia AI Services with high_service

Youku Technology

Jul 5, 2022 · Artificial Intelligence

Enlarging the Long-time Dependencies via RL-based Memory Network in Movie Affective Analysis

The paper introduces a reinforcement‑learning‑driven memory network that stores and updates historical video information via DDPG, overcoming LSTM/Transformer limitations on long‑duration movie sequences, and achieves state‑of‑the‑art affective prediction on LIRIS‑ACCEDE and related datasets, with real‑world deployments in AI content inspection and film‑element knowledge graphs.

Memory Networklong-term dependenciesmovie affective analysis

0 likes · 5 min read

Enlarging the Long-time Dependencies via RL-based Memory Network in Movie Affective Analysis

Tencent Advertising Technology

Jun 25, 2021 · Artificial Intelligence

Tencent Advertising Algorithm Competition: AI-Powered Creative Engine Live Stream

The Tencent Advertising Algorithm Competition launches a live stream series on advanced AI techniques, featuring NVIDIA's Merlin recommendation system and Tencent's ad recall system, with the upcoming June 25th session focusing on AI-driven creative engines for advertising.

AI advertisingRecommendation systemsTencent Advertising

0 likes · 3 min read

Tencent Advertising Algorithm Competition: AI-Powered Creative Engine Live Stream