Computer Vision Engineer, Neural ISP & Depth

About the company

Efference provides high-performance perception and compute infrastructure for training and deploying robot models in the cloud. Our vision is to make robotic deployments scalable with shared compute.

We approach the problem by owning the entire path from the image sensor to the datacenter GPU. We shrink the load at the sensor, fix the uplink so robot traffic is fast and reliable, and clock synchronize the system so observations arrive time-aligned, ready for inference.

About the role

You'll develop neural image-processing and depth models that run on Efference devices and improve robot perception. You'll own the work from training data and model design through NPU deployment, checking that gains in image quality or depth accuracy hold up in real robotic tasks.

What you’ll do

• Develop neural and hybrid ISP pipelines for RAW processing, denoising, demosaicing, HDR, and other imaging problems where learning improves the system.
• Build monocular, stereo, or multi-camera depth models with explicit evaluation of scale, boundary quality, temporal consistency, and uncertainty.
• Construct training and validation datasets, loss functions, and ablations that expose behavior under motion, low light, reflective surfaces, and unfamiliar scenes.
• Optimize architectures, quantization, operator support, memory use, and data movement for our NPU; profile performance on the actual device.
• Measure geometric fidelity and failure modes, including invented detail or unstable depth, with image-quality, VIO, and robotics evaluation engineers.
• Integrate models into versioned camera pipelines and maintain repeatable comparisons of task performance, runtime, power, and bandwidth.

What we are looking for

• Experience developing deep-learning models for image restoration, computational imaging, stereo, depth estimation, or related geometric vision tasks.
• Strong Python and PyTorch skills, with experience building training pipelines and running controlled experiments.
• Understanding of image formation, camera geometry, and the classical processing methods relevant to your models.
• Experience optimizing or deploying models under real compute, memory, or latency constraints.
• Ability to analyze failure cases and distinguish a better benchmark score from an improvement that matters in a deployed perception system.

Nice to Have

• Direct experience with RAW/Bayer data, ISP tuning, multi-view calibration, or temporal video models.
• Work on NPU toolchains, quantization-aware training, custom operators, or hardware-aware architecture design.
• Research or production work in robot perception, learned compression, or joint optimization of imaging and downstream tasks.