三维视觉算法实习生（图像 / 视频美型）

优缺点分析

角色解读

• Progress from algorithm intern to full-time researcher/engineer working on core technology.
• Opportunity to publish research papers at top conferences (CVPR, ICCV, etc.).
• Transition to leading projects in AIGC for video generation and 3D content creation.
• Test and evaluate existing CNN-based beautification methods on real product data.
• Fine-tune and improve deep learning models for 2D/3D human tracking and reconstruction.
• Design and implement novel network modules for local deformation and body shape correction.
• Develop algorithm demos and collaborate with product teams to drive deployment.
• Strong Python and PyTorch skills for deep learning model development.
• Understanding of 3D vision fundamentals, including SMPL, human pose, and reconstruction.
• Ability to read and reproduce research papers independently.
• Familiarity with AI coding tools and good engineering practices.

申请策略

• Demonstrate deep curiosity about the intersection of AI and aesthetics in your cover letter.
• Mention specific problems you would like to solve in 3D human beautification.
• Emphasize any projects involving 3D reconstruction, human pose estimation, or SMPL.
• Showcase contributions to open-source projects or research publications in related fields.
• Highlight strong coding skills with PyTorch and experience in training deep models.
• Strengthen understanding of diffusion models and video generation techniques.
• Practice implementing papers from scratch to build reproducibility skills.

面试指南

• Use the STAR method (Situation, Task, Action, Result) for project-based questions.
• Structure technical answers by first stating the problem, then your approach, and finally trade-offs.
• Show enthusiasm by connecting your past experiences to the company's mission.
• Explain how you would approach 3D human mesh recovery from a single image.
• Describe a project where you fine-tuned a deep learning model
• what challenges did you face?
• What are the key components of a diffusion model for image generation?
• How do you ensure temporal consistency in video-based human tracking?