Description
We are seeking a passionate researcher-engineer who is interested in the 3D direction of machine learning. You will be involved in the full cycle: from researching new architectures in papers to implementing them into working pipelines, turning fresh scientific ideas into working solutions for generating and improving 3D objects.
Responsibilities
- Development, training, and validation of diffusion models for 3D generation, reconstruction, and segmentation tasks.
- Preparation and creation of large-scale 3D datasets: working with synthetic data, processing meshes and point clouds, rendering (Blender, PyTorch3D).
- Research work: analyzing the latest scientific papers (CVPR, ICCV, NeurIPS, SIGGRAPH) and implementing promising approaches.
- Optimizing models for efficient inference (including work with TensorRT).
- Solving tasks related to 3D generation: retopology, splitting meshes into components, UV parameterization.
Requirements
- You have experience solving CV tasks and practical experience training diffusion models.
- Deep understanding of architectures: Transformers, U-Net, principles of multimodal conditioning (text, image, 3D), LoRA, adapters.
- Proficient in Python and proficient in the deep learning stack: PyTorch, Diffusers, Hugging Face, have experience with 3D libraries (PyTorch3D, Open3D, trimesh).
A big plus would be:
- A portfolio or pet projects demonstrating your interest in the field (3D generation, NeRF, Gaussian Splatting, 3D segmentation tasks, etc.).
- Skills in Blender (Python scripting) or similar tools for creating synthetic data.
- Understanding of the fundamentals of computer graphics (rendering, materials, lighting).
- Experience optimizing models using TensorRT, ONNX, or similar tools.
Conditions
- Working on a non-standard task at the intersection of machine learning, computer vision, and three-dimensional geometry.
- Working with complex and interesting data: from stylized game assets to parametric 3D models.
- Showcasing work results at professional events: from digital visualizations to presentations of printed generated objects (e.g., AIJ, Moscow's Video Game Week, etc.).
- Comfortable modern office near Kutuzovskaya metro station
- Hybrid or remote work
- Annual bonus
- Corporate gym and relaxation areas
- Extended DMS + family insurance
- Employee mortgage more favorable (-1/3 of the current interest rate)
- Free SberPrime+ subscription, discounts on products from partner companies
- Reward for recommending friends to the Sber team.