Role overview
We are looking for an ML Engineer with hands-on image upscaling (single-image super-resolution) experience to build and ship production-quality upscaling and artifact suppression models for real-time / interactive visual applications. The work combines computer vision, perceptual quality optimization, and GPU-focused deployment, with integration into graphics stacks and strong engineering practices (reproducibility, CI/CD, benchmarking, telemetry).
What you'll work on
Develop and improve single-image super-resolution and artifact suppression models (noise, compression artifacts, ringing, halos)
Optimize quality using a balanced approach between PSNR/SSIM and perceptual quality (e.g., LPIPS)
Build efficient training and data pipelines (incl. mixed precision, distributed training when needed)
Deploy and optimize models for GPU / edge inference (latency, memory, throughput)
Integrate ML components into graphics/visual pipelines (coordination with graphics/engine teams)
Run rigorous benchmarking and maintain model quality/perf dashboards; document design decisions and trade-offs
Collaborate cross-functionally with graphics, product, QA, and platform teams
What we're looking for
5+ years in ML / Computer Vision, with experience shipping production image models for real-time or interactive use
Practical experience in single-image SR (upscaling) and artifact suppression
Strong understanding of perceptual vs distortion metrics trade-offs (PSNR/SSIM vs perceptual metrics like LPIPS)
Proficiency in PyTorch (or equivalent) and strong Python
Model deployment/optimization experience on GPUs: TensorRT or ONNX Runtime plus profiling/latency/memory tuning
Strong engineering discipline: clean code, testing basics, version control, reproducible experiments
One systems language: C++ preferred (or similar)
Color and image processing fundamentals: sRGB vs linear, YUV, HDR, resampling/filtering basics (bicubic/Lanczos, anti-aliasing)
Familiarity with graphics pipelines and shaders; experience integrating ML into graphics stacks
CUDA/cuDNN hands-on optimization experience
Experience with model serving, versioning, telemetry, A/B testing
Experience with standard SR datasets/benchmarks (DIV2K, Set5/14, BSD100, Urban100, Manga109)
Experience with distributed training at scale and production-grade data pipelines