https://chensjtu.github.io/Chen Yang — Research and Product Updates2026-08-31T00:00:00+08:00Recent research, publications, and 3D intelligence product updates from Chen Yang.Chen Yanghttps://orcid.org/0000-0003-4496-7849https://chensjtu.github.io/papers/sculpt.htmlSCULPT: Subtractive Composition for 3D Part Generation2026-08-13T00:00:00+08:002026-08-13T00:00:00+08:00Chen Yang is a co-first author and provided research guidance on SCULPT, which introduces subtractive composition in a structured 3D latent space to generate adaptive, editable, part-aware 3D assets with consistent shared boundaries.https://chensjtu.github.io/#news-pixel-artisan-pro-usersPixel Artisan PRO surpasses 40,000 users2026-06-12T00:00:00+08:002026-06-12T00:00:00+08:00The Pixel Artisan PRO team product surpassed 40,000 users within 50 hours of launch, bringing 3D creation capabilities to a broad user community.https://chensjtu.github.io/#news-xiaoyi3d-pixel-artisan-proXiaoyi 3D and Pixel Artisan PRO launch at HDC 20262026-06-12T00:00:00+08:002026-06-12T00:00:00+08:00Chen Yang owned the model R&D delivered into Xiaoyi 3D and Pixel Artisan PRO, while Jiemin Fang was responsible for overall program coordination and progress management. The team product launched at HDC 2026 and received the HarmonyOS Innovative Premium App Award.https://chensjtu.github.io/#news-snap-snapSnap-Snap accepted to TCSVT2026-06-01T00:00:00+08:002026-06-01T00:00:00+08:00Snap-Snap was accepted to IEEE Transactions on Circuits and Systems for Video Technology.https://sjtu-deepvisionlab.github.io/WorldAct/WorldAct: Activating Monolithic 3D Worlds into Interactive-Ready Object-Centric Scenes2026-05-15T00:00:00+08:002026-05-15T00:00:00+08:00Chen Yang is a co-author and proposed the idea on WorldAct, which transforms static monolithic 3D Gaussian worlds into editable, object-centric scenes that support interaction and embodied tasks.https://chensjtu.github.io/papers/tigon.htmlTIGON: Text-Image Conditioned 3D Generation2026-03-22T00:00:00+08:002026-03-22T00:00:00+08:00TIGON uses text-image conditions and multimodal guidance to improve controllable 3D generation while preserving geometric and appearance quality.https://chensjtu.github.io/#news-tigon-cvpr-2026TIGON accepted to CVPR 20262026-03-22T00:00:00+08:002026-03-22T00:00:00+08:00TIGON was accepted to the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026.https://chensjtu.github.io/#news-unilat3d-cvpr-findings-2026UniLat3D accepted to Findings of CVPR 20262026-03-01T00:00:00+08:002026-03-01T00:00:00+08:00UniLat3D was accepted to Findings of CVPR 2026 for its unified geometry-appearance latent representation and single-stage 3D generation framework.https://chensjtu.github.io/#news-worldgrow-galleryWorldGrow Gallery goes online2026-01-15T00:00:00+08:002026-01-15T00:00:00+08:00The WorldGrow Gallery is now online, showcasing examples of coherent, infinitely extendable 3D world generation.https://chensjtu.github.io/#news-worldgrow-aaai-oralWorldGrow accepted as an Oral presentation at AAAI 20262025-11-15T00:00:00+08:002025-11-15T00:00:00+08:00WorldGrow was accepted as an Oral presentation at the AAAI Conference on Artificial Intelligence 2026.https://chensjtu.github.io/papers/worldgrow.htmlWorldGrow: Generating Infinite 3D World2025-10-24T00:00:00+08:002025-10-24T00:00:00+08:00WorldGrow extends coherent 3D scenes block by block using structured 3D priors, context-aware inpainting, and coarse-to-fine synthesis for open-ended world generation.https://chensjtu.github.io/papers/unilat3d.htmlUniLat3D: Geometry-Appearance Unified Latents for Single-Stage 3D Generation2025-09-29T00:00:00+08:002025-09-29T00:00:00+08:00UniLat3D unifies geometry and appearance in a shared latent representation, enabling single-stage 3D generation with a simple and efficient generative process.https://chensjtu.github.io/#news-gaussianobject-starsGaussianObject surpasses 1,200 GitHub stars2025-07-15T00:00:00+08:002025-07-15T00:00:00+08:00GaussianObject surpassed 1,200 GitHub stars, reflecting continued interest in sparse-view 3D object reconstruction with Gaussian splatting.https://chensjtu.github.io/papers/gaussianobject.htmlGaussianObject: High-Quality 3D Object Reconstruction from Four Views with Gaussian Splatting2024-02-15T00:00:00+08:002024-02-15T00:00:00+08:00GaussianObject reconstructs high-quality 3D objects from four views with Gaussian splatting, geometric initialization, floater elimination, and diffusion-based completion.