Domain-Specific 3D Reconstruction
project on efficient machine learning model query for scientific visualization
Photorealistic 3D Antler Reconstruction (Few-Image NeRF): We’re building a domain-specialized NeRF system that turns one or a few trail-camera photos into an accurate, photorealistic 3D antler model for virtual scoring—enabling non-invasive, pre-harvest measurement to support ethical hunting and wildlife management. Trained on a curated dataset of high-resolution 3D scans, our method couples mesh-aware geometric supervision with retrieval-augmented inference (pose estimation + nearest-neighbor priors) to produce robust results from sparse views, culminating in an easy-to-use prototype for field and industry use.