DepthEvidence 4B Model Integrates Metric Depth for Spatial Reasoning

DepthEvidence is a 4B parameter model that uses a camera-conditioned decoder to predict full-resolution metric depth. It introduces a dense-to-language interface that converts depth predictions into object-aligned continuous geometry tokens for improved spatial and numerical reasoning.