Sr. Data Scientist – Urban Scene Generation

Overview

Passionate about 3D deep learning research and solutions that will impact the work of hundreds of thousands of users? Join our AI Prototypes Team and push the frontiers of artificial intelligence working with point clouds, photogrammetric meshes, oriented imagery and elevation rasters to create new and exciting tools for 3D feature extraction and generation. This is a great opportunity to join a research-oriented team that works with the SOTA deep learning architectures and ideas, publishes its research results, and contributes to development of next-gen Esri software.

Responsibilities

• Research and implementation of generative deep learning architectures, training of neural network models for the Urban Scene Generator (USG) project – AI-assisted Urban Planning and Urban Design: LOD1 and LOD2+ buildings, street furniture, terrain, orthophoto, vegetation, and more
• Work with client-app teams’ stakeholders (CityEngine, ArcGIS Urban, Professional Services) on alignment of the R&D goals and market needs, trajectories
• Contribute to neural network design and architectural decisions, do actual implementations, run local and distributed trainings, code automated hyperparameter search
• Work on deployment pipeline:
  • Scalable and robust Inference-as-a-Service ArcGIS Online offering (USG service)
  • MLOps, Continuous Integration, Continuous Deployment
  • Service federation & Billing
• Work with the data engineering team on defining the specs for training and inference datasets
• Stay current on latest SOTA developments in generative neural networks across multiple domains: raster, meshes, point clouds, LLM, and more
• Collaborate with other researchers and developers within and across the teams throughout the R&D process
• Be ready to advise on design and architectural decisions outside of the USG project
• Solve and articulate complex problems with design, rapid prototypes, and user experiences
• Additional R&D duties as assigned

Requirements

• Demonstrated 5+ years of combined experience with the following:
  • GANs in general, Pix2PixHD-alike architectures in particular
  • Transformers
  • Diffusion architectures
  • Multimodal foundation models
  • PyTorch, PyTorch Lightning, Distributed Data Parallel, Distributed Model Parallel
  • 3D formats and structures: Wavefront OBJ, Pytorch3d or Kaolin meshes and camera models
  • Ubuntu, Docker, AWS cloud (EC2 compute)
  • ArcGIS Server REST API, Geoprocessing services, ArcGIS Pro, CityEngine
  • SCRUM
• Familiarity with the domains of Urban Planning and Urban Design
• Strong research background
• Excellent written and verbal communication skills
• Bachelor’s in computer or data science, mathematics, robotics, remote sensing, computer vision, AI, computational design

Recommended Qualifications

• Experience with C++, CGAL
• Familiarity with Differentiable Rendering, Reinforcement Learning
• Experience with CityEngine, CGA and procedural content generation, ArcGIS Network Analyst
• Experience with Autodesk Revit, Rhino Grasshopper, BIM
• Understanding of scalable CPU- or GPU-based processing of large volumes of point clouds and rasters
• Experience with NVIDIA RAPIDS, NVIDIA Kaolin, PyTorch Geometric
• Master’s or Ph.D. in computer or data science, mathematics, robotics, remote sensing, computer vision, AI, computational design

#LI-RC2

#LI-Remote