A 17‑Year‑Old High‑Schooler Becomes First‑Author on a CVPR Paper

CVPR 2026 received 16,092 full papers out of over 30,000 submissions, with an acceptance rate of 25.42%. Among the accepted papers, Hu Jin‑cheng, a 17‑year‑old high‑school senior from Anhui Ansheng School, is the sole first‑author from a Chinese high school.

The accepted work, titled CraftMesh: High‑Fidelity Generative Mesh Manipulation via Poisson Seamless Fusion , was developed under the guidance of Prof. Li‑gang Liu and Dr. You‑cheng Cai at the University of Science and Technology of China (USTC) GCL Lab, which had 11 papers accepted at CVPR this year.

CraftMesh addresses the challenge of controllable, high‑fidelity editing of explicit 3D meshes. Its pipeline decomposes the task into three stages: (1) 2D image editing, (2) generation of a target‑region mesh, and (3) Poisson‑based seamless fusion that blends the generated mesh with the original model using a hybrid SDF/mesh representation.

Technically, direct vertex optimization often leads to unstable convergence and noise. CraftMesh mitigates this by employing Poisson normal blending to guide indirect SDF optimization, preserving topology while achieving smooth geometric transitions. For texture, the framework uses distribution‑aware color alignment, gradient‑preserving Poisson fusion, and smooth‑transition constraints to retain high‑frequency details.

Quantitative evaluations show that CraftMesh outperforms baseline methods on CLIP similarity and other metrics, demonstrating better global structural consistency and detail preservation in tasks such as mesh part insertion, deletion, and deformation.

The system supports both text‑driven commands and drag‑based spatial interactions, making it flexible for various 3D content creation scenarios in gaming, AR/VR, and digital manufacturing.

While high‑school participation in top conferences is not unprecedented—three Chinese high‑school teams earned Spotlights at NeurIPS 2024 in the High School Projects Track—Hu’s paper was submitted to the regular CVPR main track and underwent the same double‑blind review as all other submissions. Both cases highlight the role of leading university labs in mentoring young talent.

The article concludes that the ability of teenagers to read cutting‑edge literature, conduct experiments, and produce conference‑level papers reflects the democratization of AI research through open‑source tools and widespread compute, but also warns against treating conference papers as shortcuts to success, emphasizing the importance of genuine curiosity and rigorous scientific practice.