美图3D Vision Algorithm Intern (3D Human Reconstruction & Tracking)

任职要求

Experience
·      PhD candidate in Computer Science, Artificial Intelligence, Intelligent Arts, or related fields
·      Proficient in Python, with experience using common libraries (e.g., NumPy, Transformers, Diffusers, Open3D), and strong hands-on experience with deep learning frameworks such as PyTorch
·      Strong, sustained interest in image/video aesthetics or 3D art; resilient mindset and openness to change

Skills & Abilities
·      Solid engineering implementation skills and rigorous experimental practices; able to independently complete module-level tasks
·      Proficient with AI-assisted programming tools; maintains clean, well-structured code
·      Strong ability to read, analyse, summarize, and reproduce academic papers
·      Clear logical thinking, strong problem-decomposition skills, and curiosity about underlying principles

Motivation & Values
·      Long-term interest in AI + imaging / 3D / aesthetics
·      Proactive, self-driven, and willing to explore new approaches and challenges
·      Alignment with a product culture centred on user experience and technical value

Bonus Points
·      Prior experience with SMPL, human pose estimation, 3D reconstruction, or tracking
·      Familiarity with video understanding, diffusion models, autoregressive models, or flow-matching methods
·      Research projects, comp…

工作职责

Location & Duration
Sydney Central; 6-12 months


Role Overview
You will participate in the research and development of human aesthetic enhancement and spatiotemporally consistent editing technologies at Meitu. You will work directly with real, product-scale datasets and state-of-the-art algorithms.
Depending on the internship track, your work may include (but is not limited to):
·      Fine-grained and controllable image / video aesthetic enhancement
·      2D / 3D human tracking and 3D reconstruction
·      Regression, reconstruction, and structural constraints of digital human models (e.g., SMPL)

This role offers the opportunity to produce both production-ready technical outcomes and high-quality academic research results. It is a research-and-engineering-oriented internship, ideal for candidates with strong interest and capability in 3D vision fundamentals, human visual quality enhancement, video generation models, and 3D human modelling.


Key Responsibilities
·      Research and implement algorithms related to depth estimation, multi-view generation, and 2D / 3D tracking with spatiotemporal reconstruction
·      Follow state-of-the-art 3D vision papers and open-source projects; reproduce experiments and adapt methods to practical applications
·      Collaborate with data teams to refine the 3D aesthetic development pipeline, improve data collection and quality evaluation, and establish foundations for high-quality scaling
·      Explore the integration of human structure priors (Skeleton / SMPL / Mesh) with multi-modal cues such as depth, normals, and optical flow in reconstruction and generative models
·      Assist in building data processing, evaluation, and visualization tools (e.g., immersive video aesthetic editing) to support rapid iteration
·      Enable high-quality projection of 3D features into 2D visual outputs, with the goal of producing A-level or above academic publications