AI Research Paper Accepted & Published at ECCV 2024
We are thrilled to announce that our research paper "Scalable Group Choreography via Variational Phase Manifold Learning" has been accepted and published at the European Conference on Computer Vision @eccvconf 2024!
This research, presented by the AIOZ AI Team in Milan, led to the development of a generative model that solves the scalability challenge in group dance generation while preserving naturalness and synchronization.
Let's further analyze this research below:
Existing group dance generation models typically face limitations in scalability. They either generate dances for a fixed number of performers or suffer from high memory consumption due to architectural constraints.
These approaches are restricted to a predetermined number of dancers. They often based on collaborative mechanisms such as cross-entity or global attention and severely limiting their application in real-world scenarios requiring larger group choreographies.
To address these challenges, we proposed a novel approach to scalable group dance generation using a phase-based variational generative model known as Phase-conditioned Dance VAE (PDVAE).
Our method achieves high-fidelity group dance motion and enables the generation of realistic and synchronized group dance performances without increasing computational and memory costs, even as the number of dancers grows.
The intensive experiments we carried out on two public datasets prove that our proposed method outperforms recent state-of-the-art approaches by a large margin and is scalable to accommodate a greater number of dancers beyond the training data.
Based on this, we can safely conclude that PDVAE provides a flexible and efficient solution for generating crowd-scale dance animations, with potential applications in diverse fields such as entertainment, virtual reality, education, and media production.
If you would like to dive deeper into this research work, you can refer to the links below:
▪️ Paper: https://www.ecva.net/papers/eccv_2024/papers_ECCV/papers/02766.pdf
▪️ Project Page: https://aioz-ai.github.io/VAE/
▪️ Poster: https://eccv2024.ecva.net/virtual/2024/poster/1394
About the AIOZ Network
AIOZ Network is a DePIN for AIOZ AI, AIOZ Storage, and AIOZ Stream.
AIOZ empowers a fast, secure, and decentralized future.
Powered by a global community of AIOZ DePIN, AIOZ rewards you for sharing your computational resources for storing, transcoding, and streaming digital media content and powering decentralized AI computation.
Find Us
AIOZ All Links | Website | Twitter | Telegram
Related Content
Lightweight Text Generation with SmolLM-135M: Fast, Compact, and Capable
Text generation remains one of the most widely used AI capabilities. From drafting articles and composing captions to structuring short narratives and writing stories, creators and builders are constantly seeking models that can deliver high-quality text with minimal computational resources. SmolLM-135M introduces compact and efficient text generation that makes high-quality language synthesis more accessible and practical for real-world applications. About SmolLM-135M SmolLM-135M is a light
Archer Image Generator: Authentic Archer-Style Artwork Made Simple
Now available on AIOZ AI—the collaborative marketplace powered by AIOZ DePIN—Archer Image Generator is a specialized text-to-image model designed to produce illustrations with sharp lines, flat shading, and the punchy, animated look fans associate with the TV show Archer. Trained on screenshots from the series alongside AI-generated images and user-contributed content, it captures the show’s unique look and feel by including “Archer style” tokens in your prompts. Whether you’re a fan of the ser
Bring Your Photos To Life With Cartoonize Image Diffusion
Now available on AIOZ AI V1—the collaborative marketplace powered by AIOZ DePIN—the Cartoonize Image Diffusion model transforms real photos into vibrant, stylized cartoons using simple, natural-language instructions. This customized diffusion model builds on Stable Diffusion 1.5 with instruction-tuning techniques from FLAN and the conditional editing approach of InstructPix2Pix, enabling direct & high-fidelity cartoonization without per-image fine-tuning. It excels at interpreting textual promp
XFeat: Accelerated Features for Lightweight Image Matching
Now available on AIOZ AI—the collaborative marketplace powered by AIOZ DePIN—the XFeat model delivers fast, lightweight, and accurate feature detection and matching for images captured from different viewpoints. Designed for efficiency, XFeat extracts discriminative keypoints and descriptors before performing rapid correspondence matching. This method makes it well-suited for resource-constrained environments where speed and reliability matter. Hosted on AIOZ AI using the PyTorch framework, XF
Color Harmonization: Smarter Color Control For More Coherent Designs
Now available on AIOZ AI—the collaborative marketplace powered by AIOZ DePIN—the Color Harmonization model transforms images by adjusting and enhancing color balance according to harmony principles, creating visually captivating and aesthetically balanced compositions. This computational model applies selected harmony templates to align colors, improving coherence while preserving details and visual impact. Based on the work of Amir Hossein Kargaran and implemented in PyTorch, it excels in ima
Git over SSH And Git LFS: A Major Update on AIOZ AI
AIOZ AI is rolling out a powerful new capability: full support for Git over SSH (Secure Shell) with Git LFS (Large File Storage). Developers and creators can now manage source code and large AI assets - model weights, datasets, media files - directly on the AIOZ AI platform with speed, security, and zero friction. This is version control built for modern AI workflows. Why This Update Matters AI projects are large, complex, and resource-heavy. Traditional Git isn’t built for multi-gigabyte fi