Research engineer in 3D reconstruction and neural rendering

The Inria Centre at Rennes University is one of Inria's eight centres and has more than thirty research teams. The Inria Centre is a major and recognized player in the field of digital sciences. It is at the heart of a rich R&D and innovation ecosystem: highly innovative PMEs, large industrial groups, competitiveness clusters, research and higher education players, laboratories of excellence, technological research institute, etc.

In recent years, there has been a surge in novel neural shape and radiance representations for reconstructing and modeling 3D scenes and objects within deep learning frameworks. These advancements enable both novel view synthesis and 3D shape recovery from images. The representations span implicit and explicit forms. Implicit representations learn spatially conditioned neural fields, such as volume density in NeRF [1] or signed distance functions in NeuS [2], which are typically trained through differentiable volumetric rendering. The latest explicit representations model the scene as a collection of primitives, like Gaussian splatting, which is trained via splatting-based alpha compositing. Efficient versions of these models rely on CUDA implementations (e.g. Instant-NGP [3], Gaussian Splatting [4]), allowing for faster training and rendering.

Building on our recent efforts in self-supervised 3D reconstruction models (e.g. [5,6]), the goal of this project is to advance existing research by developing CUDA-enabled neural 3D models that can effectively learn both shape and radiance. Key challenges include learning from sparse input images, handling noisy camera poses, and pursuing physically decomposable rendering—i.e., decoupling radiance into illumination and intrinsic material properties.

[1] NeRF: Representing scenes as neural radiance fields for view synthesis. ECCV 2020
[2] Neus: Learning neural implicit surfaces by volume rendering for multi-view reconstruction. NeurIPS 2020
[3] Instant neural graphics primitives with a multiresolution hash encoding. SIGGRAPH 2022
[4] 3D Gaussian Splatting for Real-Time Radiance Field Rendering. SIGGRAPH 2023
[5] Few-Shot Unsupervised Implicit Neural Shape Representation Learning with Spatial Adversaries. ICML 2024
[6] Unsupervised Occupancy Learning from Sparse Point Cloud. CVPR 2024

The engineer’s responsibilities will include:

• Examining and benchmarking state-of-the-art methods, including NeRFs, NeuS, and Gaussian Splatting.
• Contributing improvements to these models, focusing on robust learning from sparse images and noisy camera poses, and enhancing geometric reconstruction quality.
• Extending the models to generalize to inverse rendering tasks.

Candidates should have a M.Sc. or PhD in a computer science related field. A solid background in applied mathematics, computer vision, computer graphics and machines learning, and proficiency in Python and C++ (CUDA) are required.

• • Subsidized meals
  • Partial reimbursement of public transport costs
  • Possibility of teleworking (90 days per year) and flexible organization of working hours
  • Partial payment of insurance costs

Monthly gross salary from 2 695 euros according to diploma and experience