HY3D-Bench opens 252k filtered 3D objects for generation training
- Source
- x.com
- Date
🚀 We are excited to open-source HY3D-Bench, a unified, high-quality dataset for 3D asset generation. Addressing data scarcity and inconsistent evaluation, we provide a massive, training-ready library. Core Highlights: -252k+ High-fidelity 3D objects (rigorously filtered) -240k+ Structured part-level decomposed data for fine-grained control -125k AIGC-synthesized assets for class balance -Hunyuan3D-2.1-Small: A lightweight, SOTA baseline for easy reproduction HY3D-Bench simplifies access to robust data resources, aiming to catalyze innovation across 3D perception, robotics, and digital content creation. 🔗 GitHub: https://t.co/Ba7L5ZaKJJ 🤗 Hugging Face: https://t.co/aCcMWzkU5x 📂 Dataset: https://t.co/Z3DGqFFJT5 📄 Paper:
Context
Tencent's HY3D-Bench supplies prepared 3D training resources at two levels: complete objects and separate components. The full-object subset contains more than 252,000 watertight meshes with multi-view renderings. The part-level subset covers more than 240,000 objects with segmentation labels and separate part meshes. Those representations let researchers train on whole shapes or on how components fit together. Dataset repository.
The team also provides a smaller baseline generation model trained on the full-object data. Pairing data with a baseline gives researchers a common starting point for testing changes to their own 3D systems. Release details.
articleExplorationBench: Feedback and Self-Designed Probes Drive Agent Discovery
postHy Image3.5 preview is live. 🚀 Professional-grade image generation, +30% win…
postTencent Hunyuan's EvolveScaler trips up frontier models on evolving facts
postTencent Hunyuan open-sources AuK, a unified speech generation and editing model
Checking sign-in…
Loading comments…