vchitect

Professional software vendor delivering innovative solutions on the Softono platform. Specialized in both open-source and proprietary software development.

Visit Website

Total Products

Software by vchitect

Open Source

VBench

VBench is a comprehensive benchmark suite for evaluating video generative models, recognized as a CVPR 2024 Highlight. It provides unified implementations to assess video generation capabilities across a wide spectrum of dimensions, including subject consistency, background consistency, temporal flickering, motion smoothness, aesthetic quality, imaging quality, object class adherence, color harmony, and prompt alignment. The framework leverages human-aligned evaluation methods to ensure results correlate closely with human preferences. VBench supports various video generation models and offers tools for prompt suites, sampled video analysis, and detailed metric computation. The suite includes updated versions such as VBench++ and VBench 2.0 to address evolving evaluation needs and expand covered capabilities. Researchers and developers utilize VBench to compare model performance, identify weaknesses, and guide improvements in video synthesis technology. The project offers leaderboard results on Hugging Face,

ML Frameworks

1.7K Github Stars

Open Source

SEINE

SEINE is a short-to-long video diffusion model presented at ICLR 2024 designed for high-quality generative video tasks. It addresses the challenge of extending brief input clips into coherent, temporally consistent longer videos while maintaining visual fidelity. The model excels at two primary functions: generative transition, which seamlessly interpolates between two provided video frames or short clips, and long-term prediction, which extrapolates future frames from a short input sequence. SEINE leverages a diffusion-based architecture to synthesize realistic dynamics and textures over extended durations, overcoming common issues like motion blur and conceptual drift often found in longer video synthesis. By focusing on the transition from short observations to long-range predictions, SEINE enables applications in video editing, content creation, and simulation where predicting future events or bridging temporal gaps with plausible media is required. Its approach ensures that the generated content remains

AI & Machine Learning ML Frameworks

967 Github Stars