Technical Report · July 2026
Geometry-Enhanced Portion Estimation for Multimodal LLMs
Multimodal LLMs recognize a wide range of foods zero-shot in real-world photos, yet they are weak at portion estimation. We present a small geometry-enhanced network on a frozen DINOv2 backbone with a structured softmax-ownership volume that reasons jointly over all detected foods — no depth sensor and no MLLM fine-tuning. Across three real-world benchmarks, it cuts per-food portion error by 33–41% relative to the MLLM alone, outperforms every flagship MLLM's direct estimates, and surpasses each benchmark's originally published image-only model at its own reported metric.
