AI
LeRF: Learning Reference Coordinate Frames for Perspective Taking Reasoning
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 04:38:22 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 01:18:24 UTC.
Briefings
LeRF is a framework for Vision-Language Models (VLMs) that enables perspective-taking reasoning by learning to construct explicit reference coordinate frames. The system decides when a frame is needed, predicts its origin and entity-centered orientation, and overlays it onto images for reasoning without external 3D reconstruction. Training involves supervised fine-tuning for tool invocation and reinforcement learning on spatial VQA pairs. Performance improves over backbones and outperforms existing open-source methods on diverse benchmarks.
Key points
- LeRF trains VLMs to predict explicit reference coordinate frames for viewpoint-dependent spatial reasoning, deciding when a frame is necessary before reasoning.
- The framework achieves strong performance on diverse perspective-taking benchmarks, improving over backbones and surpassing existing open-source methods.
Context and limitations
Performance is evaluated on diverse perspective-taking benchmarks; specific limitations on edge cases or real-world deployment are not detailed in the evidence.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 04:38:22 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiewoaea4v66bdiqz36ljyw4f2oessj7zcs3agw53adnbshu3csgee